Whamcloud - gitweb
43265ab74273a570c11d8fcdd7fc2aaabeb59327
[fs/lustre-release.git] / 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, 2017, 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 #include <linux/user_namespace.h>
41 #include <linux/uidgid.h>
42
43 #define DEBUG_SUBSYSTEM S_LLITE
44
45 #include <obd_support.h>
46 #include <lustre_fid.h>
47 #include <lustre_dlm.h>
48 #include "llite_internal.h"
49
50 static int ll_create_it(struct inode *dir, struct dentry *dentry,
51                         struct lookup_intent *it,
52                         void *secctx, __u32 secctxlen);
53
54 /* called from iget5_locked->find_inode() under inode_lock spinlock */
55 static int ll_test_inode(struct inode *inode, void *opaque)
56 {
57         struct ll_inode_info    *lli = ll_i2info(inode);
58         struct lustre_md        *md = opaque;
59
60         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
61                 CERROR("MDS body missing FID\n");
62                 return 0;
63         }
64
65         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
66                 return 0;
67
68         return 1;
69 }
70
71 static int ll_set_inode(struct inode *inode, void *opaque)
72 {
73         struct ll_inode_info *lli = ll_i2info(inode);
74         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
75
76         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
77                 CERROR("MDS body missing FID\n");
78                 return -EINVAL;
79         }
80
81         lli->lli_fid = body->mbo_fid1;
82         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
83                 CERROR("Can not initialize inode "DFID" without object type: "
84                        "valid = %#llx\n",
85                        PFID(&lli->lli_fid), body->mbo_valid);
86                 return -EINVAL;
87         }
88
89         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
90         if (unlikely(inode->i_mode == 0)) {
91                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
92                 return -EINVAL;
93         }
94
95         ll_lli_init(lli);
96
97         return 0;
98 }
99
100
101 /**
102  * Get an inode by inode number(@hash), which is already instantiated by
103  * the intent lookup).
104  */
105 struct inode *ll_iget(struct super_block *sb, ino_t hash,
106                       struct lustre_md *md)
107 {
108         struct inode    *inode;
109         int             rc = 0;
110
111         ENTRY;
112
113         LASSERT(hash != 0);
114         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
115         if (inode == NULL)
116                 RETURN(ERR_PTR(-ENOMEM));
117
118         if (inode->i_state & I_NEW) {
119                 rc = ll_read_inode2(inode, md);
120                 if (rc == 0 && S_ISREG(inode->i_mode) &&
121                     ll_i2info(inode)->lli_clob == NULL)
122                         rc = cl_file_inode_init(inode, md);
123
124                 if (rc != 0) {
125                         /* Let's clear directory lsm here, otherwise
126                          * make_bad_inode() will reset the inode mode
127                          * to regular, then ll_clear_inode will not
128                          * be able to clear lsm_md */
129                         if (S_ISDIR(inode->i_mode))
130                                 ll_dir_clear_lsm_md(inode);
131                         make_bad_inode(inode);
132                         unlock_new_inode(inode);
133                         iput(inode);
134                         inode = ERR_PTR(rc);
135                 } else {
136                         inode_has_no_xattr(inode);
137                         unlock_new_inode(inode);
138                 }
139         } else if (is_bad_inode(inode)) {
140                 iput(inode);
141                 inode = ERR_PTR(-ESTALE);
142         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
143                 rc = ll_update_inode(inode, md);
144                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
145                        PFID(&md->body->mbo_fid1), inode, rc);
146                 if (rc != 0) {
147                         if (S_ISDIR(inode->i_mode))
148                                 ll_dir_clear_lsm_md(inode);
149                         iput(inode);
150                         inode = ERR_PTR(rc);
151                 }
152         }
153
154         RETURN(inode);
155 }
156
157 static void ll_invalidate_negative_children(struct inode *dir)
158 {
159         struct dentry *dentry, *tmp_subdir;
160         DECLARE_LL_D_HLIST_NODE_PTR(p);
161
162         spin_lock(&dir->i_lock);
163         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
164                 spin_lock(&dentry->d_lock);
165                 if (!list_empty(&dentry->d_subdirs)) {
166                         struct dentry *child;
167
168                         list_for_each_entry_safe(child, tmp_subdir,
169                                                  &dentry->d_subdirs,
170                                                  d_child) {
171                                 if (child->d_inode == NULL)
172                                         d_lustre_invalidate(child, 1);
173                         }
174                 }
175                 spin_unlock(&dentry->d_lock);
176         }
177         spin_unlock(&dir->i_lock);
178 }
179
180 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
181 {
182         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
183 }
184
185 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
186 {
187         struct lu_env *env;
188         struct ll_inode_info *lli = ll_i2info(inode);
189         __u16 refcheck;
190         int rc;
191         ENTRY;
192
193         if (!lli->lli_clob) {
194                 /* due to DoM read on open, there may exist pages for Lustre
195                  * regular file even though cl_object is not set up yet. */
196                 truncate_inode_pages(inode->i_mapping, 0);
197                 RETURN(0);
198         }
199
200         env = cl_env_get(&refcheck);
201         if (IS_ERR(env))
202                 RETURN(PTR_ERR(env));
203
204         /* reach MDC layer to flush data under  the DoM ldlm lock */
205         rc = cl_object_flush(env, lli->lli_clob, lock);
206         if (rc == -ENODATA) {
207                 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
208                        PFID(ll_inode2fid(inode)));
209                 /* most likely result of layout change, do nothing */
210                 rc = 0;
211         }
212
213         cl_env_put(env, &refcheck);
214         RETURN(rc);
215 }
216
217 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
218 {
219         struct inode *inode = ll_inode_from_resource_lock(lock);
220         struct ll_inode_info *lli;
221         __u64 bits = to_cancel;
222         int rc;
223
224         ENTRY;
225
226         if (!inode) {
227                 /* That means the inode is evicted most likely and may cause
228                  * the skipping of lock cleanups below, so print the message
229                  * about that in log.
230                  */
231                 if (lock->l_resource->lr_lvb_inode)
232                         LDLM_DEBUG(lock,
233                                    "can't take inode for the lock (%sevicted)\n",
234                                    lock->l_resource->lr_lvb_inode->i_state &
235                                    I_FREEING ? "" : "not ");
236                 RETURN_EXIT;
237         }
238
239         if (!fid_res_name_eq(ll_inode2fid(inode),
240                              &lock->l_resource->lr_name)) {
241                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
242                            PFID(ll_inode2fid(inode)), inode);
243                 LBUG();
244         }
245
246         if (bits & MDS_INODELOCK_XATTR) {
247                 ll_xattr_cache_destroy(inode);
248                 bits &= ~MDS_INODELOCK_XATTR;
249         }
250
251         /* For OPEN locks we differentiate between lock modes
252          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
253         if (bits & MDS_INODELOCK_OPEN)
254                 ll_have_md_lock(inode, &bits, lock->l_req_mode);
255
256         if (bits & MDS_INODELOCK_OPEN) {
257                 fmode_t fmode;
258
259                 switch (lock->l_req_mode) {
260                 case LCK_CW:
261                         fmode = FMODE_WRITE;
262                         break;
263                 case LCK_PR:
264                         fmode = FMODE_EXEC;
265                         break;
266                 case LCK_CR:
267                         fmode = FMODE_READ;
268                         break;
269                 default:
270                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
271                         LBUG();
272                 }
273
274                 ll_md_real_close(inode, fmode);
275
276                 bits &= ~MDS_INODELOCK_OPEN;
277         }
278
279         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
280                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
281                     MDS_INODELOCK_DOM))
282                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
283
284         if (bits & MDS_INODELOCK_DOM) {
285                 rc =  ll_dom_lock_cancel(inode, lock);
286                 if (rc < 0)
287                         CDEBUG(D_INODE, "cannot flush DoM data "
288                                DFID": rc = %d\n",
289                                PFID(ll_inode2fid(inode)), rc);
290         }
291
292         if (bits & MDS_INODELOCK_LAYOUT) {
293                 struct cl_object_conf conf = {
294                         .coc_opc = OBJECT_CONF_INVALIDATE,
295                         .coc_inode = inode,
296                 };
297
298                 rc = ll_layout_conf(inode, &conf);
299                 if (rc < 0)
300                         CDEBUG(D_INODE, "cannot invalidate layout of "
301                                DFID": rc = %d\n",
302                                PFID(ll_inode2fid(inode)), rc);
303         }
304
305         lli = ll_i2info(inode);
306
307         if (bits & MDS_INODELOCK_UPDATE)
308                 ll_file_set_flag(lli, LLIF_UPDATE_ATIME);
309
310         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
311                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
312                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
313                        lli, PFID(&lli->lli_pfid));
314                 truncate_inode_pages(inode->i_mapping, 0);
315
316                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
317                         struct inode *master_inode = NULL;
318                         unsigned long hash;
319
320                         /* This is slave inode, since all of the child dentry
321                          * is connected on the master inode, so we have to
322                          * invalidate the negative children on master inode */
323                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
324                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
325
326                         hash = cl_fid_build_ino(&lli->lli_pfid,
327                                         ll_need_32bit_api(ll_i2sbi(inode)));
328
329                         /* Do not lookup the inode with ilookup5, otherwise
330                          * it will cause dead lock,
331                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
332                          * it enqueues master and all of its slaves lock,
333                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
334                          * master and stripe0 lock, it will send the enqueue
335                          * req (for stripe1) to MDT1, then MDT1 finds the lock
336                          * has been granted to client2. Then MDT1 sends blocking
337                          * ast to client2.
338                          * 2. At the same time, client2 tries to unlink
339                          * the striped dir (rm -rf striped_dir), and during
340                          * lookup, it will hold the master inode of the striped
341                          * directory, whose inode state is NEW, then tries to
342                          * revalidate all of its slaves, (ll_prep_inode()->
343                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
344                          * And it will be blocked on the server side because
345                          * of 1.
346                          * 3. Then the client get the blocking_ast req, cancel
347                          * the lock, but being blocked if using ->ilookup5()),
348                          * because master inode state is NEW. */
349                         master_inode = ilookup5_nowait(inode->i_sb, hash,
350                                                         ll_test_inode_by_fid,
351                                                         (void *)&lli->lli_pfid);
352                         if (master_inode) {
353                                 ll_invalidate_negative_children(master_inode);
354                                 iput(master_inode);
355                         }
356                 } else {
357                         ll_invalidate_negative_children(inode);
358                 }
359         }
360
361         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
362             inode->i_sb->s_root != NULL &&
363             inode != inode->i_sb->s_root->d_inode)
364                 ll_invalidate_aliases(inode);
365
366         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
367                 forget_all_cached_acls(inode);
368
369         iput(inode);
370         RETURN_EXIT;
371 }
372
373 /* Check if the given lock may be downgraded instead of canceling and
374  * that convert is really needed. */
375 int ll_md_need_convert(struct ldlm_lock *lock)
376 {
377         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
378         struct inode *inode;
379         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
380         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
381         enum ldlm_mode mode = LCK_MINMODE;
382
383         if (!lock->l_conn_export ||
384             !exp_connect_lock_convert(lock->l_conn_export))
385                 return 0;
386
387         if (!wanted || !bits || ldlm_is_cancel(lock))
388                 return 0;
389
390         /* do not convert locks other than DOM for now */
391         if (!((bits | wanted) & MDS_INODELOCK_DOM))
392                 return 0;
393
394         /* We may have already remaining bits in some other lock so
395          * lock convert will leave us just extra lock for the same bit.
396          * Check if client has other lock with the same bits and the same
397          * or lower mode and don't convert if any.
398          */
399         switch (lock->l_req_mode) {
400         case LCK_PR:
401                 mode = LCK_PR;
402                 /* fallthrough */
403         case LCK_PW:
404                 mode |= LCK_CR;
405                 break;
406         case LCK_CW:
407                 mode = LCK_CW;
408                 /* fallthrough */
409         case LCK_CR:
410                 mode |= LCK_CR;
411                 break;
412         default:
413                 /* do not convert other modes */
414                 return 0;
415         }
416
417         /* is lock is too old to be converted? */
418         lock_res_and_lock(lock);
419         if (ktime_after(ktime_get(),
420                         ktime_add(lock->l_last_used,
421                                   ktime_set(ns->ns_dirty_age_limit, 0)))) {
422                 unlock_res_and_lock(lock);
423                 return 0;
424         }
425         unlock_res_and_lock(lock);
426
427         inode = ll_inode_from_resource_lock(lock);
428         ll_have_md_lock(inode, &bits, mode);
429         iput(inode);
430         return !!(bits);
431 }
432
433 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
434                        void *data, int flag)
435 {
436         struct lustre_handle lockh;
437         int rc;
438
439         ENTRY;
440
441         switch (flag) {
442         case LDLM_CB_BLOCKING:
443         {
444                 __u64 cancel_flags = LCF_ASYNC;
445
446                 /* if lock convert is not needed then still have to
447                  * pass lock via ldlm_cli_convert() to keep all states
448                  * correct, set cancel_bits to full lock bits to cause
449                  * full cancel to happen.
450                  */
451                 if (!ll_md_need_convert(lock)) {
452                         lock_res_and_lock(lock);
453                         lock->l_policy_data.l_inodebits.cancel_bits =
454                                         lock->l_policy_data.l_inodebits.bits;
455                         unlock_res_and_lock(lock);
456                 }
457                 rc = ldlm_cli_convert(lock, cancel_flags);
458                 if (!rc)
459                         RETURN(0);
460                 /* continue with cancel otherwise */
461                 ldlm_lock2handle(lock, &lockh);
462                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
463                 if (rc < 0) {
464                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
465                         RETURN(rc);
466                 }
467                 break;
468         }
469         case LDLM_CB_CANCELING:
470         {
471                 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
472
473                 /* Nothing to do for non-granted locks */
474                 if (!ldlm_is_granted(lock))
475                         break;
476
477                 /* If 'ld' is supplied then bits to be cancelled are passed
478                  * implicitly by lock converting and cancel_bits from 'ld'
479                  * should be used. Otherwise full cancel is being performed
480                  * and lock inodebits are used.
481                  *
482                  * Note: we cannot rely on cancel_bits in lock itself at this
483                  * moment because they can be changed by concurrent thread,
484                  * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
485                  * in 'ld' parameter.
486                  */
487                 if (ld) {
488                         /* partial bits cancel allowed only during convert */
489                         LASSERT(ldlm_is_converting(lock));
490                         /* mask cancel bits by lock bits so only no any unused
491                          * bits are passed to ll_lock_cancel_bits()
492                          */
493                         to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
494                 }
495                 ll_lock_cancel_bits(lock, to_cancel);
496                 break;
497         }
498         default:
499                 LBUG();
500         }
501
502         RETURN(0);
503 }
504
505 __u32 ll_i2suppgid(struct inode *i)
506 {
507         if (in_group_p(i->i_gid))
508                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
509         else
510                 return (__u32) __kgid_val(INVALID_GID);
511 }
512
513 /* Pack the required supplementary groups into the supplied groups array.
514  * If we don't need to use the groups from the target inode(s) then we
515  * instead pack one or more groups from the user's supplementary group
516  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
517 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
518 {
519         LASSERT(i1 != NULL);
520         LASSERT(suppgids != NULL);
521
522         suppgids[0] = ll_i2suppgid(i1);
523
524         if (i2)
525                 suppgids[1] = ll_i2suppgid(i2);
526         else
527                 suppgids[1] = -1;
528 }
529
530 /*
531  * try to reuse three types of dentry:
532  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
533  *    by concurrent .revalidate).
534  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
535  *    be cleared by others calling d_lustre_revalidate).
536  * 3. DISCONNECTED alias.
537  */
538 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
539 {
540         struct dentry *alias, *discon_alias, *invalid_alias;
541         DECLARE_LL_D_HLIST_NODE_PTR(p);
542
543         if (ll_d_hlist_empty(&inode->i_dentry))
544                 return NULL;
545
546         discon_alias = invalid_alias = NULL;
547
548         spin_lock(&inode->i_lock);
549         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
550                 LASSERT(alias != dentry);
551
552                 spin_lock(&alias->d_lock);
553                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
554                     S_ISDIR(inode->i_mode))
555                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
556                         discon_alias = alias;
557                 else if (alias->d_parent == dentry->d_parent             &&
558                          alias->d_name.hash == dentry->d_name.hash       &&
559                          alias->d_name.len == dentry->d_name.len         &&
560                          memcmp(alias->d_name.name, dentry->d_name.name,
561                                 dentry->d_name.len) == 0)
562                         invalid_alias = alias;
563                 spin_unlock(&alias->d_lock);
564
565                 if (invalid_alias)
566                         break;
567         }
568         alias = invalid_alias ?: discon_alias ?: NULL;
569         if (alias) {
570                 spin_lock(&alias->d_lock);
571                 dget_dlock(alias);
572                 spin_unlock(&alias->d_lock);
573         }
574         spin_unlock(&inode->i_lock);
575
576         return alias;
577 }
578
579 /*
580  * Similar to d_splice_alias(), but lustre treats invalid alias
581  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
582  */
583 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
584 {
585         struct dentry *new;
586         int rc;
587
588         if (inode) {
589                 new = ll_find_alias(inode, de);
590                 if (new) {
591                         rc = ll_d_init(new);
592                         if (rc < 0) {
593                                 dput(new);
594                                 return ERR_PTR(rc);
595                         }
596                         d_move(new, de);
597                         iput(inode);
598                         CDEBUG(D_DENTRY,
599                                "Reuse dentry %p inode %p refc %d flags %#x\n",
600                               new, new->d_inode, ll_d_count(new), new->d_flags);
601                         return new;
602                 }
603         }
604         rc = ll_d_init(de);
605         if (rc < 0)
606                 return ERR_PTR(rc);
607         d_add(de, inode);
608         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
609                de, de->d_inode, ll_d_count(de), de->d_flags);
610         return de;
611 }
612
613 static int ll_lookup_it_finish(struct ptlrpc_request *request,
614                                struct lookup_intent *it,
615                                struct inode *parent, struct dentry **de,
616                                void *secctx, __u32 secctxlen)
617 {
618         struct inode             *inode = NULL;
619         __u64                     bits = 0;
620         int                       rc;
621         struct dentry *alias;
622         ENTRY;
623
624         /* NB 1 request reference will be taken away by ll_intent_lock()
625          * when I return */
626         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
627                it->it_disposition);
628         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
629                 struct req_capsule *pill = &request->rq_pill;
630                 struct mdt_body *body = req_capsule_server_get(pill,
631                                                                &RMF_MDT_BODY);
632
633                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
634                 if (rc)
635                         RETURN(rc);
636
637                 if (it->it_op & IT_OPEN)
638                         ll_dom_finish_open(inode, request, it);
639
640                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
641
642                 /* We used to query real size from OSTs here, but actually
643                  * this is not needed. For stat() calls size would be updated
644                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
645                  * 2.4 and
646                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
647                  * Everybody else who needs correct file size would call
648                  * ll_glimpse_size or some equivalent themselves anyway.
649                  * Also see bug 7198.
650                  */
651
652                 /* If security context was returned by MDT, put it in
653                  * inode now to save an extra getxattr from security hooks,
654                  * and avoid deadlock.
655                  */
656                 if (body->mbo_valid & OBD_MD_SECCTX) {
657                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
658                         secctxlen = req_capsule_get_size(pill,
659                                                            &RMF_FILE_SECCTX,
660                                                            RCL_SERVER);
661
662                         if (secctxlen)
663                                 CDEBUG(D_SEC, "server returned security context"
664                                        " for "DFID"\n",
665                                        PFID(ll_inode2fid(inode)));
666                 }
667
668                 if (secctx != NULL && secctxlen != 0) {
669                         inode_lock(inode);
670                         rc = security_inode_notifysecctx(inode, secctx,
671                                                          secctxlen);
672                         inode_unlock(inode);
673                         if (rc)
674                                 CWARN("cannot set security context for "
675                                       DFID": rc = %d\n",
676                                       PFID(ll_inode2fid(inode)), rc);
677                 }
678         }
679
680         /* Only hash *de if it is unhashed (new dentry).
681          * Atoimc_open may passin hashed dentries for open.
682          */
683         alias = ll_splice_alias(inode, *de);
684         if (IS_ERR(alias))
685                 GOTO(out, rc = PTR_ERR(alias));
686
687         *de = alias;
688
689         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
690                 /* we have lookup look - unhide dentry */
691                 if (bits & MDS_INODELOCK_LOOKUP)
692                         d_lustre_revalidate(*de);
693         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
694                 /*
695                  * If file was created on the server, the dentry is revalidated
696                  * in ll_create_it if the lock allows for it.
697                  */
698                 /* Check that parent has UPDATE lock. */
699                 struct lookup_intent parent_it = {
700                                         .it_op = IT_GETATTR,
701                                         .it_lock_handle = 0 };
702                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
703
704                 /* If it is striped directory, get the real stripe parent */
705                 if (unlikely(ll_dir_striped(parent))) {
706                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
707                                                  ll_i2info(parent)->lli_lsm_md,
708                                                  (*de)->d_name.name,
709                                                  (*de)->d_name.len, &fid);
710                         if (rc != 0)
711                                 GOTO(out, rc);
712                 }
713
714                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
715                                        NULL)) {
716                         d_lustre_revalidate(*de);
717                         ll_intent_release(&parent_it);
718                 }
719         }
720
721         GOTO(out, rc = 0);
722
723 out:
724         if (rc != 0 && it->it_op & IT_OPEN) {
725                 ll_intent_drop_lock(it);
726                 ll_open_cleanup((*de)->d_sb, request);
727         }
728
729         return rc;
730 }
731
732 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
733                                    struct lookup_intent *it,
734                                    void **secctx, __u32 *secctxlen,
735                                    struct pcc_create_attach *pca)
736 {
737         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
738         struct dentry *save = dentry, *retval;
739         struct ptlrpc_request *req = NULL;
740         struct md_op_data *op_data = NULL;
741         struct lov_user_md *lum = NULL;
742         __u32 opc;
743         int rc;
744         char secctx_name[XATTR_NAME_MAX + 1];
745
746         ENTRY;
747
748         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
749                 RETURN(ERR_PTR(-ENAMETOOLONG));
750
751         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
752                dentry->d_name.len, dentry->d_name.name,
753                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
754
755         if (d_mountpoint(dentry))
756                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
757
758         if (it == NULL || it->it_op == IT_GETXATTR)
759                 it = &lookup_it;
760
761         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
762                 rc = ll_statahead(parent, &dentry, 0);
763                 if (rc == 1)
764                         RETURN(dentry == save ? NULL : dentry);
765         }
766
767         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
768             dentry->d_sb->s_flags & SB_RDONLY)
769                 RETURN(ERR_PTR(-EROFS));
770
771         if (it->it_op & IT_CREAT)
772                 opc = LUSTRE_OPC_CREATE;
773         else
774                 opc = LUSTRE_OPC_ANY;
775
776         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
777                                      dentry->d_name.len, 0, opc, NULL);
778         if (IS_ERR(op_data))
779                 GOTO(out, retval = ERR_CAST(op_data));
780
781         /* enforce umask if acl disabled or MDS doesn't support umask */
782         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
783                 it->it_create_mode &= ~current_umask();
784
785         if (it->it_op & IT_CREAT &&
786             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
787                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
788                                              &dentry->d_name,
789                                              &op_data->op_file_secctx_name,
790                                              &op_data->op_file_secctx,
791                                              &op_data->op_file_secctx_size);
792                 if (rc < 0)
793                         GOTO(out, retval = ERR_PTR(rc));
794                 if (secctx != NULL)
795                         *secctx = op_data->op_file_secctx;
796                 if (secctxlen != NULL)
797                         *secctxlen = op_data->op_file_secctx_size;
798         } else {
799                 if (secctx != NULL)
800                         *secctx = NULL;
801                 if (secctxlen != NULL)
802                         *secctxlen = 0;
803         }
804
805         /* ask for security context upon intent */
806         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
807                 /* get name of security xattr to request to server */
808                 rc = ll_listsecurity(parent, secctx_name,
809                                      sizeof(secctx_name));
810                 if (rc < 0) {
811                         CDEBUG(D_SEC, "cannot get security xattr name for "
812                                DFID": rc = %d\n",
813                                PFID(ll_inode2fid(parent)), rc);
814                 } else if (rc > 0) {
815                         op_data->op_file_secctx_name = secctx_name;
816                         op_data->op_file_secctx_name_size = rc;
817                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
818                                rc, secctx_name, PFID(ll_inode2fid(parent)));
819                 }
820         }
821
822         if (pca && pca->pca_dataset) {
823                 struct pcc_dataset *dataset = pca->pca_dataset;
824
825                 OBD_ALLOC_PTR(lum);
826                 if (lum == NULL)
827                         GOTO(out, retval = ERR_PTR(-ENOMEM));
828
829                 lum->lmm_magic = LOV_USER_MAGIC_V1;
830                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
831                 op_data->op_data = lum;
832                 op_data->op_data_size = sizeof(*lum);
833                 op_data->op_archive_id = dataset->pccd_rwid;
834
835                 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
836                                    op_data);
837                 if (rc)
838                         GOTO(out, retval = ERR_PTR(rc));
839
840                 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
841                                       &pca->pca_dentry);
842                 if (rc)
843                         GOTO(out, retval = ERR_PTR(rc));
844
845                 it->it_flags |= MDS_OPEN_PCC;
846         }
847
848         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
849                             &ll_md_blocking_ast, 0);
850         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
851          * client does not know which suppgid should be sent to the MDS, or
852          * some other(s) changed the target file's GID after this RPC sent
853          * to the MDS with the suppgid as the original GID, then we should
854          * try again with right suppgid. */
855         if (rc == -EACCES && it->it_op & IT_OPEN &&
856             it_disposition(it, DISP_OPEN_DENY)) {
857                 struct mdt_body *body;
858
859                 LASSERT(req != NULL);
860
861                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
862                 if (op_data->op_suppgids[0] == body->mbo_gid ||
863                     op_data->op_suppgids[1] == body->mbo_gid ||
864                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
865                         GOTO(out, retval = ERR_PTR(-EACCES));
866
867                 fid_zero(&op_data->op_fid2);
868                 op_data->op_suppgids[1] = body->mbo_gid;
869                 ptlrpc_req_finished(req);
870                 req = NULL;
871                 ll_intent_release(it);
872                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
873                                     &ll_md_blocking_ast, 0);
874         }
875
876         if (rc < 0)
877                 GOTO(out, retval = ERR_PTR(rc));
878
879         /* dir layout may change */
880         ll_unlock_md_op_lsm(op_data);
881         rc = ll_lookup_it_finish(req, it, parent, &dentry,
882                                  secctx != NULL ? *secctx : NULL,
883                                  secctxlen != NULL ? *secctxlen : 0);
884         if (rc != 0) {
885                 ll_intent_release(it);
886                 GOTO(out, retval = ERR_PTR(rc));
887         }
888
889         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
890             !S_ISREG(dentry->d_inode->i_mode) &&
891             !S_ISDIR(dentry->d_inode->i_mode)) {
892                 ll_release_openhandle(dentry, it);
893         }
894         ll_lookup_finish_locks(it, dentry);
895
896         GOTO(out, retval = (dentry == save) ? NULL : dentry);
897
898 out:
899         if (op_data != NULL && !IS_ERR(op_data)) {
900                 if (secctx != NULL && secctxlen != NULL) {
901                         /* caller needs sec ctx info, so reset it in op_data to
902                          * prevent it from being freed */
903                         op_data->op_file_secctx = NULL;
904                         op_data->op_file_secctx_size = 0;
905                 }
906                 ll_finish_md_op_data(op_data);
907         }
908
909         if (lum != NULL)
910                 OBD_FREE_PTR(lum);
911
912         ptlrpc_req_finished(req);
913         return retval;
914 }
915
916 #ifdef HAVE_IOP_ATOMIC_OPEN
917 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
918                                    unsigned int flags)
919 {
920         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
921         struct dentry *de;
922
923         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
924                dentry->d_name.len, dentry->d_name.name,
925                PFID(ll_inode2fid(parent)), parent, flags);
926
927         /*
928          * Optimize away (CREATE && !OPEN). Let .create handle the race.
929          * but only if we have write permissions there, otherwise we need
930          * to proceed with lookup. LU-4185
931          */
932         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
933             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
934                 return NULL;
935
936         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
937                 itp = NULL;
938         else
939                 itp = &it;
940         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL);
941
942         if (itp != NULL)
943                 ll_intent_release(itp);
944
945         return de;
946 }
947
948 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
949 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
950 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
951 # define ll_last_arg
952 # define ll_set_created(o, f)                                           \
953 do {                                                                    \
954         (f)->f_mode |= FMODE_CREATED;                                   \
955 } while (0)
956
957 #else
958 # define ll_is_opened(o, f)             (*(o))
959 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
960 # define ll_last_arg                    , int *opened
961 # define ll_set_created(o, f)                                           \
962 do {                                                                    \
963         *(o) |= FILE_CREATED;                                           \
964 } while (0)
965
966 #endif
967
968 /*
969  * For cached negative dentry and new dentry, handle lookup/create/open
970  * together.
971  */
972 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
973                           struct file *file, unsigned open_flags,
974                           umode_t mode ll_last_arg)
975 {
976         struct lookup_intent *it;
977         struct dentry *de;
978         long long lookup_flags = LOOKUP_OPEN;
979         void *secctx = NULL;
980         __u32 secctxlen = 0;
981         struct ll_sb_info *sbi;
982         struct pcc_create_attach pca = { NULL, NULL };
983         int rc = 0;
984         ENTRY;
985
986         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
987                            "open_flags %x, mode %x opened %d\n",
988                dentry->d_name.len, dentry->d_name.name,
989                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
990                ll_is_opened(opened, file));
991
992         /* Only negative dentries enter here */
993         LASSERT(dentry->d_inode == NULL);
994
995         if (!d_unhashed(dentry)) {
996                 /* A valid negative dentry that just passed revalidation,
997                  * there's little point to try and open it server-side,
998                  * even though there's a minuscule chance it might succeed.
999                  * Either way it's a valid race to just return -ENOENT here.
1000                  */
1001                 if (!(open_flags & O_CREAT))
1002                         return -ENOENT;
1003
1004                 /* Otherwise we just unhash it to be rehashed afresh via
1005                  * lookup if necessary
1006                  */
1007                 d_drop(dentry);
1008         }
1009
1010         OBD_ALLOC(it, sizeof(*it));
1011         if (!it)
1012                 RETURN(-ENOMEM);
1013
1014         it->it_op = IT_OPEN;
1015         if (open_flags & O_CREAT) {
1016                 it->it_op |= IT_CREAT;
1017                 lookup_flags |= LOOKUP_CREATE;
1018                 sbi = ll_i2sbi(dir);
1019                 /* Volatile file is used for HSM restore, so do not use PCC */
1020                 if (!filename_is_volatile(dentry->d_name.name,
1021                                           dentry->d_name.len, NULL)) {
1022                         struct pcc_matcher item;
1023                         struct pcc_dataset *dataset;
1024
1025                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1026                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1027                         item.pm_projid = ll_i2info(dir)->lli_projid;
1028                         item.pm_name = &dentry->d_name;
1029                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1030                                                         &item);
1031                         pca.pca_dataset = dataset;
1032                 }
1033         }
1034         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1035         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1036         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1037
1038         /* Dentry added to dcache tree in ll_lookup_it */
1039         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca);
1040         if (IS_ERR(de))
1041                 rc = PTR_ERR(de);
1042         else if (de != NULL)
1043                 dentry = de;
1044
1045         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1046
1047         if (!rc) {
1048                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1049                         /* Dentry instantiated in ll_create_it. */
1050                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
1051                         security_release_secctx(secctx, secctxlen);
1052                         if (rc) {
1053                                 /* We dget in ll_splice_alias. */
1054                                 if (de != NULL)
1055                                         dput(de);
1056                                 goto out_release;
1057                         }
1058
1059                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1060                         if (rc) {
1061                                 if (de != NULL)
1062                                         dput(de);
1063                                 GOTO(out_release, rc);
1064                         }
1065
1066                         ll_set_created(opened, file);
1067                 } else {
1068                         /* Open the file with O_CREAT, but the file already
1069                          * existed on MDT. This may happend in the case that
1070                          * the LOOKUP ibits lock is revoked and the
1071                          * corresponding dentry cache is deleted.
1072                          * i.e. In the current Lustre, the truncate operation
1073                          * will revoke the LOOKUP ibits lock, and the file
1074                          * dentry cache will be invalidated. The following open
1075                          * with O_CREAT flag will call into ->atomic_open, the
1076                          * file was wrongly though as newly created file and
1077                          * try to auto cache the file. So after client knows it
1078                          * is not a DISP_OPEN_CREATE, it should cleanup the
1079                          * already created PCC copy.
1080                          */
1081                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1082                 }
1083
1084                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1085                         /* Open dentry. */
1086                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1087                                 /* We cannot call open here as it might
1088                                  * deadlock. This case is unreachable in
1089                                  * practice because of OBD_CONNECT_NODEVOH. */
1090                                 rc = finish_no_open(file, de);
1091                         } else {
1092                                 file->private_data = it;
1093                                 rc = ll_finish_open(file, dentry, opened);
1094                                 /* We dget in ll_splice_alias. finish_open takes
1095                                  * care of dget for fd open.
1096                                  */
1097                                 if (de != NULL)
1098                                         dput(de);
1099                         }
1100                 } else {
1101                         rc = finish_no_open(file, de);
1102                 }
1103         } else {
1104                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1105         }
1106
1107 out_release:
1108         ll_intent_release(it);
1109         OBD_FREE(it, sizeof(*it));
1110
1111         RETURN(rc);
1112 }
1113
1114 #else /* !HAVE_IOP_ATOMIC_OPEN */
1115 static struct lookup_intent *
1116 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1117 {
1118         struct lookup_intent *it;
1119
1120         OBD_ALLOC_PTR(it);
1121         if (!it)
1122                 return ERR_PTR(-ENOMEM);
1123
1124         if (lookup_flags & LOOKUP_OPEN) {
1125                 it->it_op = IT_OPEN;
1126                 /* Avoid file creation for ro bind mount point(is_readonly) */
1127                 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1128                         it->it_op |= IT_CREAT;
1129                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1130                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1131                                                 ~(is_readonly ? O_CREAT : 0));
1132                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1133         } else {
1134                 it->it_op = IT_GETATTR;
1135         }
1136
1137         return it;
1138 }
1139
1140 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1141                                    struct nameidata *nd)
1142 {
1143         struct dentry *de;
1144         ENTRY;
1145
1146         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1147                 struct lookup_intent *it;
1148
1149                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1150                         it = ll_d2d(dentry)->lld_it;
1151                         ll_d2d(dentry)->lld_it = NULL;
1152                 } else {
1153                         /*
1154                          * Optimize away (CREATE && !OPEN). Let .create handle
1155                          * the race. But only if we have write permissions
1156                          * there, otherwise we need to proceed with lookup.
1157                          * LU-4185
1158                          */
1159                         if ((nd->flags & LOOKUP_CREATE) &&
1160                             !(nd->flags & LOOKUP_OPEN) &&
1161                             (inode_permission(parent,
1162                                               MAY_WRITE | MAY_EXEC) == 0))
1163                                 RETURN(NULL);
1164
1165                         it = ll_convert_intent(&nd->intent.open, nd->flags,
1166                                 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1167                                 (nd->path.mnt->mnt_sb->s_flags & SB_RDONLY));
1168                         if (IS_ERR(it))
1169                                 RETURN((struct dentry *)it);
1170                 }
1171
1172                 de = ll_lookup_it(parent, dentry, it, NULL, NULL, NULL);
1173                 if (de)
1174                         dentry = de;
1175                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1176                         if (dentry->d_inode &&
1177                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1178                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1179                                         /* We cannot call open here as it might
1180                                          * deadlock. This case is unreachable in
1181                                          * practice because of
1182                                          * OBD_CONNECT_NODEVOH. */
1183                                 } else {
1184                                         struct file *filp;
1185
1186                                         nd->intent.open.file->private_data = it;
1187                                         filp = lookup_instantiate_filp(nd,
1188                                                                        dentry,
1189                                                                        NULL);
1190                                         if (IS_ERR(filp)) {
1191                                                 if (de)
1192                                                         dput(de);
1193                                                 de = (struct dentry *)filp;
1194                                         }
1195                                 }
1196                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1197                                 /* XXX This can only reliably work on assumption
1198                                  * that there are NO hashed negative dentries.*/
1199                                 ll_d2d(dentry)->lld_it = it;
1200                                 it = NULL; /* Will be freed in ll_create_nd */
1201                                 /* We absolutely depend on ll_create_nd to be
1202                                  * called to not leak this intent and possible
1203                                  * data attached to it */
1204                         }
1205                 }
1206
1207                 if (it) {
1208                         ll_intent_release(it);
1209                         OBD_FREE(it, sizeof(*it));
1210                 }
1211         } else {
1212                 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL, NULL);
1213         }
1214
1215         RETURN(de);
1216 }
1217 #endif /* HAVE_IOP_ATOMIC_OPEN */
1218
1219 /* We depend on "mode" being set with the proper file type/umask by now */
1220 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1221 {
1222         struct inode *inode = NULL;
1223         struct ptlrpc_request *request = NULL;
1224         struct ll_sb_info *sbi = ll_i2sbi(dir);
1225         int rc;
1226         ENTRY;
1227
1228         LASSERT(it && it->it_disposition);
1229
1230         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1231         request = it->it_request;
1232         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1233         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1234         if (rc)
1235                 GOTO(out, inode = ERR_PTR(rc));
1236
1237         /* Pause to allow for a race with concurrent access by fid */
1238         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1239
1240         /* We asked for a lock on the directory, but were granted a
1241          * lock on the inode.  Since we finally have an inode pointer,
1242          * stuff it in the lock. */
1243         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1244                PFID(ll_inode2fid(inode)), inode);
1245         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1246         EXIT;
1247  out:
1248         ptlrpc_req_finished(request);
1249         return inode;
1250 }
1251
1252 /*
1253  * By the time this is called, we already have created the directory cache
1254  * entry for the new file, but it is so far negative - it has no inode.
1255  *
1256  * We defer creating the OBD object(s) until open, to keep the intent and
1257  * non-intent code paths similar, and also because we do not have the MDS
1258  * inode number before calling ll_create_node() (which is needed for LOV),
1259  * so we would need to do yet another RPC to the MDS to store the LOV EA
1260  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1261  * lmm_size in datalen (the MDS still has code which will handle that).
1262  *
1263  * If the create succeeds, we fill in the inode information
1264  * with d_instantiate().
1265  */
1266 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1267                         struct lookup_intent *it,
1268                         void *secctx, __u32 secctxlen)
1269 {
1270         struct inode *inode;
1271         __u64 bits = 0;
1272         int rc = 0;
1273         ENTRY;
1274
1275         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1276                dentry->d_name.len, dentry->d_name.name,
1277                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1278
1279         rc = it_open_error(DISP_OPEN_CREATE, it);
1280         if (rc)
1281                 RETURN(rc);
1282
1283         inode = ll_create_node(dir, it);
1284         if (IS_ERR(inode))
1285                 RETURN(PTR_ERR(inode));
1286
1287         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1288             secctx != NULL) {
1289                 inode_lock(inode);
1290                 /* must be done before d_instantiate, because it calls
1291                  * security_d_instantiate, which means a getxattr if security
1292                  * context is not set yet */
1293                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1294                 inode_unlock(inode);
1295                 if (rc)
1296                         RETURN(rc);
1297         }
1298
1299         d_instantiate(dentry, inode);
1300
1301         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1302                 rc = ll_inode_init_security(dentry, inode, dir);
1303                 if (rc)
1304                         RETURN(rc);
1305         }
1306
1307         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1308         if (bits & MDS_INODELOCK_LOOKUP)
1309                 d_lustre_revalidate(dentry);
1310
1311         RETURN(0);
1312 }
1313
1314 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1315 {
1316         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1317                                                        &RMF_MDT_BODY);
1318
1319         LASSERT(body);
1320         if (body->mbo_valid & OBD_MD_FLMTIME &&
1321             body->mbo_mtime > inode->i_mtime.tv_sec) {
1322                 CDEBUG(D_INODE,
1323                        "setting fid " DFID " mtime from %lld to %llu\n",
1324                        PFID(ll_inode2fid(inode)),
1325                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1326                 inode->i_mtime.tv_sec = body->mbo_mtime;
1327         }
1328
1329         if (body->mbo_valid & OBD_MD_FLCTIME &&
1330             body->mbo_ctime > inode->i_ctime.tv_sec)
1331                 inode->i_ctime.tv_sec = body->mbo_ctime;
1332 }
1333
1334 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1335                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1336 {
1337         struct qstr *name = &dchild->d_name;
1338         struct ptlrpc_request *request = NULL;
1339         struct md_op_data *op_data;
1340         struct inode *inode = NULL;
1341         struct ll_sb_info *sbi = ll_i2sbi(dir);
1342         int tgt_len = 0;
1343         int err;
1344
1345         ENTRY;
1346         if (unlikely(tgt != NULL))
1347                 tgt_len = strlen(tgt) + 1;
1348
1349 again:
1350         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1351                                      name->len, 0, opc, NULL);
1352         if (IS_ERR(op_data))
1353                 GOTO(err_exit, err = PTR_ERR(op_data));
1354
1355         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1356                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1357                                               &op_data->op_file_secctx_name,
1358                                               &op_data->op_file_secctx,
1359                                               &op_data->op_file_secctx_size);
1360                 if (err < 0)
1361                         GOTO(err_exit, err);
1362         }
1363
1364         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1365                         from_kuid(&init_user_ns, current_fsuid()),
1366                         from_kgid(&init_user_ns, current_fsgid()),
1367                         cfs_curproc_cap_pack(), rdev, &request);
1368 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1369         /*
1370          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1371          * fetch default LMV here.
1372          */
1373         if (unlikely(err == -EREMOTE)) {
1374                 struct ll_inode_info    *lli = ll_i2info(dir);
1375                 struct lmv_user_md      *lum;
1376                 int                     lumsize;
1377                 int                     err2;
1378
1379                 ptlrpc_req_finished(request);
1380                 request = NULL;
1381
1382                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1383                                         OBD_MD_DEFAULT_MEA);
1384                 ll_finish_md_op_data(op_data);
1385                 op_data = NULL;
1386                 if (err2 == 0) {
1387                         struct lustre_md md = { NULL };
1388
1389                         md.body = req_capsule_server_get(&request->rq_pill,
1390                                                          &RMF_MDT_BODY);
1391                         if (!md.body)
1392                                 GOTO(err_exit, err = -EPROTO);
1393
1394                         OBD_ALLOC_PTR(md.default_lmv);
1395                         if (!md.default_lmv)
1396                                 GOTO(err_exit, err = -ENOMEM);
1397
1398                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1399                         md.default_lmv->lsm_md_stripe_count =
1400                                 lum->lum_stripe_count;
1401                         md.default_lmv->lsm_md_master_mdt_index =
1402                                 lum->lum_stripe_offset;
1403                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1404
1405                         err = ll_update_inode(dir, &md);
1406                         md_free_lustre_md(sbi->ll_md_exp, &md);
1407                         if (err)
1408                                 GOTO(err_exit, err);
1409                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1410                         /*
1411                          * If there are no default stripe EA on the MDT, but the
1412                          * client has default stripe, then it probably means
1413                          * default stripe EA has just been deleted.
1414                          */
1415                         down_write(&lli->lli_lsm_sem);
1416                         if (lli->lli_default_lsm_md)
1417                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1418                         lli->lli_default_lsm_md = NULL;
1419                         up_write(&lli->lli_lsm_sem);
1420                 } else {
1421                         GOTO(err_exit, err);
1422                 }
1423
1424                 ptlrpc_req_finished(request);
1425                 request = NULL;
1426                 goto again;
1427         }
1428 #endif
1429
1430         if (err < 0)
1431                 GOTO(err_exit, err);
1432
1433         ll_update_times(request, dir);
1434
1435         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1436
1437         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1438         if (err)
1439                 GOTO(err_exit, err);
1440
1441         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1442                 inode_lock(inode);
1443                 /* must be done before d_instantiate, because it calls
1444                  * security_d_instantiate, which means a getxattr if security
1445                  * context is not set yet */
1446                 err = security_inode_notifysecctx(inode,
1447                                                   op_data->op_file_secctx,
1448                                                   op_data->op_file_secctx_size);
1449                 inode_unlock(inode);
1450                 if (err)
1451                         GOTO(err_exit, err);
1452         }
1453
1454         d_instantiate(dchild, inode);
1455
1456         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1457                 err = ll_inode_init_security(dchild, inode, dir);
1458                 if (err)
1459                         GOTO(err_exit, err);
1460         }
1461
1462         EXIT;
1463 err_exit:
1464         if (request != NULL)
1465                 ptlrpc_req_finished(request);
1466
1467         if (!IS_ERR_OR_NULL(op_data))
1468                 ll_finish_md_op_data(op_data);
1469
1470         return err;
1471 }
1472
1473 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1474                     dev_t rdev)
1475 {
1476         struct qstr *name = &dchild->d_name;
1477         ktime_t kstart = ktime_get();
1478         int err;
1479         ENTRY;
1480
1481         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1482                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1483                mode, rdev);
1484
1485         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1486                 mode &= ~current_umask();
1487
1488         switch (mode & S_IFMT) {
1489         case 0:
1490                 mode |= S_IFREG;
1491                 /* fallthrough */
1492         case S_IFREG:
1493         case S_IFCHR:
1494         case S_IFBLK:
1495         case S_IFIFO:
1496         case S_IFSOCK:
1497                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1498                                   LUSTRE_OPC_MKNOD);
1499                 break;
1500         case S_IFDIR:
1501                 err = -EPERM;
1502                 break;
1503         default:
1504                 err = -EINVAL;
1505         }
1506
1507         if (!err)
1508                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1509                                    ktime_us_delta(ktime_get(), kstart));
1510
1511         RETURN(err);
1512 }
1513
1514 #ifdef HAVE_IOP_ATOMIC_OPEN
1515 /*
1516  * Plain create. Intent create is handled in atomic_open.
1517  */
1518 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1519                         umode_t mode, bool want_excl)
1520 {
1521         ktime_t kstart = ktime_get();
1522         int rc;
1523
1524         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1525
1526         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1527                            "flags=%u, excl=%d\n", dentry->d_name.len,
1528                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1529                dir, mode, want_excl);
1530
1531         /* Using mknod(2) to create a regular file is designed to not recognize
1532          * volatile file name, so we use ll_mknod() here. */
1533         rc = ll_mknod(dir, dentry, mode, 0);
1534
1535         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1536                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1537
1538         if (!rc)
1539                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1540                                    ktime_us_delta(ktime_get(), kstart));
1541
1542         return rc;
1543 }
1544 #else /* !HAVE_IOP_ATOMIC_OPEN */
1545 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1546                         ll_umode_t mode, struct nameidata *nd)
1547 {
1548         struct ll_dentry_data *lld = ll_d2d(dentry);
1549         struct lookup_intent *it = NULL;
1550         ktime_t kstart = ktime_get();
1551         int rc;
1552
1553         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1554
1555         if (lld != NULL)
1556                 it = lld->lld_it;
1557
1558         if (!it) {
1559                 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1560                  * so that volatile file name is recoginized.
1561                  * Mknod(2), however, is designed to not recognize volatile
1562                  * file name to avoid inode leak under orphan directory until
1563                  * MDT reboot */
1564                 return ll_new_node(dir, dentry, NULL, mode, 0,
1565                                    LUSTRE_OPC_CREATE);
1566         }
1567
1568         lld->lld_it = NULL;
1569
1570         /* Was there an error? Propagate it! */
1571         if (it->it_status) {
1572                 rc = it->it_status;
1573                 goto out;
1574         }
1575
1576         rc = ll_create_it(dir, dentry, it, NULL, 0);
1577         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1578                 struct file *filp;
1579
1580                 nd->intent.open.file->private_data = it;
1581                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1582                 if (IS_ERR(filp))
1583                         rc = PTR_ERR(filp);
1584         }
1585
1586 out:
1587         ll_intent_release(it);
1588         OBD_FREE(it, sizeof(*it));
1589
1590         if (!rc)
1591                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1592                                    ktime_us_delta(ktime_get(), kstart));
1593
1594         return rc;
1595 }
1596 #endif /* HAVE_IOP_ATOMIC_OPEN */
1597
1598 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1599                       const char *oldpath)
1600 {
1601         struct qstr *name = &dchild->d_name;
1602         ktime_t kstart = ktime_get();
1603         int err;
1604         ENTRY;
1605
1606         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1607                name->len, name->name, PFID(ll_inode2fid(dir)),
1608                dir, 3000, oldpath);
1609
1610         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1611                           LUSTRE_OPC_SYMLINK);
1612
1613         if (!err)
1614                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1615                                    ktime_us_delta(ktime_get(), kstart));
1616
1617         RETURN(err);
1618 }
1619
1620 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1621                    struct dentry *new_dentry)
1622 {
1623         struct inode *src = old_dentry->d_inode;
1624         struct qstr *name = &new_dentry->d_name;
1625         struct ll_sb_info *sbi = ll_i2sbi(dir);
1626         struct ptlrpc_request *request = NULL;
1627         struct md_op_data *op_data;
1628         ktime_t kstart = ktime_get();
1629         int err;
1630
1631         ENTRY;
1632         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1633                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1634                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1635
1636         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1637                                      0, LUSTRE_OPC_ANY, NULL);
1638         if (IS_ERR(op_data))
1639                 RETURN(PTR_ERR(op_data));
1640
1641         err = md_link(sbi->ll_md_exp, op_data, &request);
1642         ll_finish_md_op_data(op_data);
1643         if (err)
1644                 GOTO(out, err);
1645
1646         ll_update_times(request, dir);
1647         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1648                            ktime_us_delta(ktime_get(), kstart));
1649         EXIT;
1650 out:
1651         ptlrpc_req_finished(request);
1652         RETURN(err);
1653 }
1654
1655 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1656 {
1657         struct qstr *name = &dchild->d_name;
1658         ktime_t kstart = ktime_get();
1659         int err;
1660         ENTRY;
1661
1662         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1663                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1664
1665         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1666                 mode &= ~current_umask();
1667
1668         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1669
1670         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1671         if (err == 0)
1672                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1673                                    ktime_us_delta(ktime_get(), kstart));
1674
1675         RETURN(err);
1676 }
1677
1678 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1679 {
1680         struct qstr *name = &dchild->d_name;
1681         struct ptlrpc_request *request = NULL;
1682         struct md_op_data *op_data;
1683         ktime_t kstart = ktime_get();
1684         int rc;
1685
1686         ENTRY;
1687
1688         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1689                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1690
1691         if (unlikely(d_mountpoint(dchild)))
1692                 RETURN(-EBUSY);
1693
1694         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1695                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1696         if (IS_ERR(op_data))
1697                 RETURN(PTR_ERR(op_data));
1698
1699         if (dchild->d_inode != NULL)
1700                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1701
1702         op_data->op_fid2 = op_data->op_fid3;
1703         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1704         ll_finish_md_op_data(op_data);
1705         if (!rc)
1706                 ll_update_times(request, dir);
1707
1708         ptlrpc_req_finished(request);
1709         if (!rc)
1710                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1711                                    ktime_us_delta(ktime_get(), kstart));
1712         RETURN(rc);
1713 }
1714
1715 /**
1716  * Remove dir entry
1717  **/
1718 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1719 {
1720         struct ptlrpc_request *request = NULL;
1721         struct md_op_data *op_data;
1722         ktime_t kstart = ktime_get();
1723         int rc;
1724         ENTRY;
1725
1726         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1727                namelen, name, PFID(ll_inode2fid(dir)), dir);
1728
1729         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1730                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1731         if (IS_ERR(op_data))
1732                 RETURN(PTR_ERR(op_data));
1733         op_data->op_cli_flags |= CLI_RM_ENTRY;
1734         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1735         ll_finish_md_op_data(op_data);
1736         if (!rc)
1737                 ll_update_times(request, dir);
1738
1739         ptlrpc_req_finished(request);
1740         if (!rc)
1741                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1742                                    ktime_us_delta(ktime_get(), kstart));
1743         RETURN(rc);
1744 }
1745
1746 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1747 {
1748         struct qstr *name = &dchild->d_name;
1749         struct ptlrpc_request *request = NULL;
1750         struct md_op_data *op_data;
1751         struct mdt_body *body;
1752         ktime_t kstart = ktime_get();
1753         int rc;
1754
1755         ENTRY;
1756
1757         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1758                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1759
1760         /*
1761          * XXX: unlink bind mountpoint maybe call to here,
1762          * just check it as vfs_unlink does.
1763          */
1764         if (unlikely(d_mountpoint(dchild)))
1765                 RETURN(-EBUSY);
1766
1767         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1768                                      LUSTRE_OPC_ANY, NULL);
1769         if (IS_ERR(op_data))
1770                 RETURN(PTR_ERR(op_data));
1771
1772         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1773
1774         op_data->op_fid2 = op_data->op_fid3;
1775         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1776         ll_finish_md_op_data(op_data);
1777         if (rc)
1778                 GOTO(out, rc);
1779
1780         /*
1781          * The server puts attributes in on the last unlink, use them to update
1782          * the link count so the inode can be freed immediately.
1783          */
1784         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1785         if (body->mbo_valid & OBD_MD_FLNLINK)
1786                 set_nlink(dchild->d_inode, body->mbo_nlink);
1787
1788         ll_update_times(request, dir);
1789
1790 out:
1791         ptlrpc_req_finished(request);
1792         if (!rc)
1793                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1794                                    ktime_us_delta(ktime_get(), kstart));
1795         RETURN(rc);
1796 }
1797
1798 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1799                      struct inode *tgt, struct dentry *tgt_dchild
1800 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1801                      , unsigned int flags
1802 #endif
1803                      )
1804 {
1805         struct qstr *src_name = &src_dchild->d_name;
1806         struct qstr *tgt_name = &tgt_dchild->d_name;
1807         struct ptlrpc_request *request = NULL;
1808         struct ll_sb_info *sbi = ll_i2sbi(src);
1809         struct md_op_data *op_data;
1810         ktime_t kstart = ktime_get();
1811         int err;
1812         ENTRY;
1813
1814 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1815         if (flags)
1816                 return -EINVAL;
1817 #endif
1818
1819         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1820                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1821                src_name->len, src_name->name,
1822                PFID(ll_inode2fid(src)), src, tgt_name->len,
1823                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1824
1825         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1826                 RETURN(-EBUSY);
1827
1828         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1829                                      LUSTRE_OPC_ANY, NULL);
1830         if (IS_ERR(op_data))
1831                 RETURN(PTR_ERR(op_data));
1832
1833         if (src_dchild->d_inode != NULL)
1834                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1835
1836         if (tgt_dchild->d_inode != NULL)
1837                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1838
1839         err = md_rename(sbi->ll_md_exp, op_data,
1840                         src_name->name, src_name->len,
1841                         tgt_name->name, tgt_name->len, &request);
1842         ll_finish_md_op_data(op_data);
1843         if (!err) {
1844                 ll_update_times(request, src);
1845                 ll_update_times(request, tgt);
1846         }
1847
1848         ptlrpc_req_finished(request);
1849
1850         if (!err) {
1851                 d_move(src_dchild, tgt_dchild);
1852                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1853                                    ktime_us_delta(ktime_get(), kstart));
1854         }
1855
1856         RETURN(err);
1857 }
1858
1859 const struct inode_operations ll_dir_inode_operations = {
1860         .mknod          = ll_mknod,
1861 #ifdef HAVE_IOP_ATOMIC_OPEN
1862         .atomic_open    = ll_atomic_open,
1863 #endif
1864         .lookup         = ll_lookup_nd,
1865         .create         = ll_create_nd,
1866         /* We need all these non-raw things for NFSD, to not patch it. */
1867         .unlink         = ll_unlink,
1868         .mkdir          = ll_mkdir,
1869         .rmdir          = ll_rmdir,
1870         .symlink        = ll_symlink,
1871         .link           = ll_link,
1872         .rename         = ll_rename,
1873         .setattr        = ll_setattr,
1874         .getattr        = ll_getattr,
1875         .permission     = ll_inode_permission,
1876 #ifdef HAVE_IOP_XATTR
1877         .setxattr       = ll_setxattr,
1878         .getxattr       = ll_getxattr,
1879         .removexattr    = ll_removexattr,
1880 #endif
1881         .listxattr      = ll_listxattr,
1882 #ifdef HAVE_IOP_GET_ACL
1883         .get_acl        = ll_get_acl,
1884 #endif
1885 #ifdef HAVE_IOP_SET_ACL
1886         .set_acl        = ll_set_acl,
1887 #endif
1888 };
1889
1890 const struct inode_operations ll_special_inode_operations = {
1891         .setattr        = ll_setattr,
1892         .getattr        = ll_getattr,
1893         .permission     = ll_inode_permission,
1894 #ifdef HAVE_IOP_XATTR
1895         .setxattr       = ll_setxattr,
1896         .getxattr       = ll_getxattr,
1897         .removexattr    = ll_removexattr,
1898 #endif
1899         .listxattr      = ll_listxattr,
1900 #ifdef HAVE_IOP_GET_ACL
1901         .get_acl        = ll_get_acl,
1902 #endif
1903 #ifdef HAVE_IOP_SET_ACL
1904         .set_acl        = ll_set_acl,
1905 #endif
1906 };