Whamcloud - gitweb
LU-12275 sec: encryption support for DoM files
[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, bool encrypt,
53                         void *encctx, __u32 encctxlen);
54
55 /* called from iget5_locked->find_inode() under inode_lock spinlock */
56 static int ll_test_inode(struct inode *inode, void *opaque)
57 {
58         struct ll_inode_info    *lli = ll_i2info(inode);
59         struct lustre_md        *md = opaque;
60
61         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
62                 CERROR("MDS body missing FID\n");
63                 return 0;
64         }
65
66         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
67                 return 0;
68
69         return 1;
70 }
71
72 static int ll_set_inode(struct inode *inode, void *opaque)
73 {
74         struct ll_inode_info *lli = ll_i2info(inode);
75         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
76
77         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
78                 CERROR("MDS body missing FID\n");
79                 return -EINVAL;
80         }
81
82         lli->lli_fid = body->mbo_fid1;
83         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
84                 CERROR("Can not initialize inode "DFID" without object type: "
85                        "valid = %#llx\n",
86                        PFID(&lli->lli_fid), body->mbo_valid);
87                 return -EINVAL;
88         }
89
90         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
91         if (unlikely(inode->i_mode == 0)) {
92                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
93                 return -EINVAL;
94         }
95
96         ll_lli_init(lli);
97
98         return 0;
99 }
100
101
102 /**
103  * Get an inode by inode number(@hash), which is already instantiated by
104  * the intent lookup).
105  */
106 struct inode *ll_iget(struct super_block *sb, ino_t hash,
107                       struct lustre_md *md)
108 {
109         struct inode    *inode;
110         int             rc = 0;
111
112         ENTRY;
113
114         LASSERT(hash != 0);
115         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
116         if (inode == NULL)
117                 RETURN(ERR_PTR(-ENOMEM));
118
119         if (inode->i_state & I_NEW) {
120                 rc = ll_read_inode2(inode, md);
121                 if (rc == 0 && S_ISREG(inode->i_mode) &&
122                     ll_i2info(inode)->lli_clob == NULL)
123                         rc = cl_file_inode_init(inode, md);
124
125                 if (rc != 0) {
126                         /* Let's clear directory lsm here, otherwise
127                          * make_bad_inode() will reset the inode mode
128                          * to regular, then ll_clear_inode will not
129                          * be able to clear lsm_md */
130                         if (S_ISDIR(inode->i_mode))
131                                 ll_dir_clear_lsm_md(inode);
132                         make_bad_inode(inode);
133                         unlock_new_inode(inode);
134                         iput(inode);
135                         inode = ERR_PTR(rc);
136                 } else {
137                         inode_has_no_xattr(inode);
138                         unlock_new_inode(inode);
139                 }
140         } else if (is_bad_inode(inode)) {
141                 iput(inode);
142                 inode = ERR_PTR(-ESTALE);
143         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
144                 rc = ll_update_inode(inode, md);
145                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
146                        PFID(&md->body->mbo_fid1), inode, rc);
147                 if (rc != 0) {
148                         if (S_ISDIR(inode->i_mode))
149                                 ll_dir_clear_lsm_md(inode);
150                         iput(inode);
151                         inode = ERR_PTR(rc);
152                 }
153         }
154
155         RETURN(inode);
156 }
157
158 static void ll_invalidate_negative_children(struct inode *dir)
159 {
160         struct dentry *dentry, *tmp_subdir;
161
162         spin_lock(&dir->i_lock);
163         hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
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, ns->ns_dirty_age_limit))) {
421                 unlock_res_and_lock(lock);
422                 return 0;
423         }
424         unlock_res_and_lock(lock);
425
426         inode = ll_inode_from_resource_lock(lock);
427         ll_have_md_lock(inode, &bits, mode);
428         iput(inode);
429         return !!(bits);
430 }
431
432 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
433                        void *data, int flag)
434 {
435         struct lustre_handle lockh;
436         int rc;
437
438         ENTRY;
439
440         switch (flag) {
441         case LDLM_CB_BLOCKING:
442         {
443                 __u64 cancel_flags = LCF_ASYNC;
444
445                 /* if lock convert is not needed then still have to
446                  * pass lock via ldlm_cli_convert() to keep all states
447                  * correct, set cancel_bits to full lock bits to cause
448                  * full cancel to happen.
449                  */
450                 if (!ll_md_need_convert(lock)) {
451                         lock_res_and_lock(lock);
452                         lock->l_policy_data.l_inodebits.cancel_bits =
453                                         lock->l_policy_data.l_inodebits.bits;
454                         unlock_res_and_lock(lock);
455                 }
456                 rc = ldlm_cli_convert(lock, cancel_flags);
457                 if (!rc)
458                         RETURN(0);
459                 /* continue with cancel otherwise */
460                 ldlm_lock2handle(lock, &lockh);
461                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
462                 if (rc < 0) {
463                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
464                         RETURN(rc);
465                 }
466                 break;
467         }
468         case LDLM_CB_CANCELING:
469         {
470                 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
471
472                 /* Nothing to do for non-granted locks */
473                 if (!ldlm_is_granted(lock))
474                         break;
475
476                 /* If 'ld' is supplied then bits to be cancelled are passed
477                  * implicitly by lock converting and cancel_bits from 'ld'
478                  * should be used. Otherwise full cancel is being performed
479                  * and lock inodebits are used.
480                  *
481                  * Note: we cannot rely on cancel_bits in lock itself at this
482                  * moment because they can be changed by concurrent thread,
483                  * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
484                  * in 'ld' parameter.
485                  */
486                 if (ld) {
487                         /* partial bits cancel allowed only during convert */
488                         LASSERT(ldlm_is_converting(lock));
489                         /* mask cancel bits by lock bits so only no any unused
490                          * bits are passed to ll_lock_cancel_bits()
491                          */
492                         to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
493                 }
494                 ll_lock_cancel_bits(lock, to_cancel);
495                 break;
496         }
497         default:
498                 LBUG();
499         }
500
501         RETURN(0);
502 }
503
504 __u32 ll_i2suppgid(struct inode *i)
505 {
506         if (in_group_p(i->i_gid))
507                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
508         else
509                 return (__u32) __kgid_val(INVALID_GID);
510 }
511
512 /* Pack the required supplementary groups into the supplied groups array.
513  * If we don't need to use the groups from the target inode(s) then we
514  * instead pack one or more groups from the user's supplementary group
515  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
516 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
517 {
518         LASSERT(i1 != NULL);
519         LASSERT(suppgids != NULL);
520
521         suppgids[0] = ll_i2suppgid(i1);
522
523         if (i2)
524                 suppgids[1] = ll_i2suppgid(i2);
525         else
526                 suppgids[1] = -1;
527 }
528
529 /*
530  * try to reuse three types of dentry:
531  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
532  *    by concurrent .revalidate).
533  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
534  *    be cleared by others calling d_lustre_revalidate).
535  * 3. DISCONNECTED alias.
536  */
537 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
538 {
539         struct dentry *alias, *discon_alias, *invalid_alias;
540
541         if (hlist_empty(&inode->i_dentry))
542                 return NULL;
543
544         discon_alias = invalid_alias = NULL;
545
546         spin_lock(&inode->i_lock);
547         hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
548                 LASSERT(alias != dentry);
549
550                 spin_lock(&alias->d_lock);
551                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
552                     S_ISDIR(inode->i_mode))
553                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
554                         discon_alias = alias;
555                 else if (alias->d_parent == dentry->d_parent             &&
556                          alias->d_name.hash == dentry->d_name.hash       &&
557                          alias->d_name.len == dentry->d_name.len         &&
558                          memcmp(alias->d_name.name, dentry->d_name.name,
559                                 dentry->d_name.len) == 0)
560                         invalid_alias = alias;
561                 spin_unlock(&alias->d_lock);
562
563                 if (invalid_alias)
564                         break;
565         }
566         alias = invalid_alias ?: discon_alias ?: NULL;
567         if (alias) {
568                 spin_lock(&alias->d_lock);
569                 dget_dlock(alias);
570                 spin_unlock(&alias->d_lock);
571         }
572         spin_unlock(&inode->i_lock);
573
574         return alias;
575 }
576
577 /*
578  * Similar to d_splice_alias(), but lustre treats invalid alias
579  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
580  */
581 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
582 {
583         struct dentry *new;
584         int rc;
585
586         if (inode) {
587                 new = ll_find_alias(inode, de);
588                 if (new) {
589                         rc = ll_d_init(new);
590                         if (rc < 0) {
591                                 dput(new);
592                                 return ERR_PTR(rc);
593                         }
594                         d_move(new, de);
595                         iput(inode);
596                         CDEBUG(D_DENTRY,
597                                "Reuse dentry %p inode %p refc %d flags %#x\n",
598                               new, new->d_inode, ll_d_count(new), new->d_flags);
599                         return new;
600                 }
601         }
602         rc = ll_d_init(de);
603         if (rc < 0)
604                 return ERR_PTR(rc);
605         d_add(de, inode);
606         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
607                de, de->d_inode, ll_d_count(de), de->d_flags);
608         return de;
609 }
610
611 static int ll_lookup_it_finish(struct ptlrpc_request *request,
612                                struct lookup_intent *it,
613                                struct inode *parent, struct dentry **de,
614                                void *secctx, __u32 secctxlen,
615                                void *encctx, __u32 encctxlen,
616                                ktime_t kstart, bool encrypt)
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 encryption context was returned by MDT, put it in
638                  * inode now to save an extra getxattr and avoid deadlock.
639                  */
640                 if (body->mbo_valid & OBD_MD_ENCCTX) {
641                         encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
642                         encctxlen = req_capsule_get_size(pill,
643                                                          &RMF_FILE_ENCCTX,
644                                                          RCL_SERVER);
645
646                         if (encctxlen) {
647                                 CDEBUG(D_SEC,
648                                        "server returned encryption ctx for "DFID"\n",
649                                        PFID(ll_inode2fid(inode)));
650                                 rc = ll_xattr_cache_insert(inode,
651                                                LL_XATTR_NAME_ENCRYPTION_CONTEXT,
652                                                            encctx, encctxlen);
653                                 if (rc)
654                                         CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
655                                               ll_i2sbi(inode)->ll_fsname,
656                                               PFID(ll_inode2fid(inode)), rc);
657                                 else if (encrypt) {
658                                         rc = llcrypt_get_encryption_info(inode);
659                                         if (rc)
660                                                 CDEBUG(D_SEC,
661                                                  "cannot get enc info for "DFID": rc = %d\n",
662                                                  PFID(ll_inode2fid(inode)), rc);
663                                 }
664                         }
665                 }
666
667                 if (it->it_op & IT_OPEN)
668                         ll_dom_finish_open(inode, request, it);
669
670                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
671
672                 /* We used to query real size from OSTs here, but actually
673                  * this is not needed. For stat() calls size would be updated
674                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
675                  * 2.4 and
676                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
677                  * Everybody else who needs correct file size would call
678                  * ll_glimpse_size or some equivalent themselves anyway.
679                  * Also see bug 7198.
680                  */
681
682                 /* If security context was returned by MDT, put it in
683                  * inode now to save an extra getxattr from security hooks,
684                  * and avoid deadlock.
685                  */
686                 if (body->mbo_valid & OBD_MD_SECCTX) {
687                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
688                         secctxlen = req_capsule_get_size(pill,
689                                                            &RMF_FILE_SECCTX,
690                                                            RCL_SERVER);
691
692                         if (secctxlen)
693                                 CDEBUG(D_SEC, "server returned security context"
694                                        " for "DFID"\n",
695                                        PFID(ll_inode2fid(inode)));
696                 }
697
698                 if (secctx != NULL && secctxlen != 0) {
699                         /* no need to protect selinux_inode_setsecurity() by
700                          * inode_lock. Taking it would lead to a client deadlock
701                          * LU-13617
702                          */
703                         rc = security_inode_notifysecctx(inode, secctx,
704                                                          secctxlen);
705                         if (rc)
706                                 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
707                                       ll_i2sbi(inode)->ll_fsname,
708                                       PFID(ll_inode2fid(inode)),
709                                       rc);
710                 }
711         }
712
713         /* Only hash *de if it is unhashed (new dentry).
714          * Atoimc_open may passin hashed dentries for open.
715          */
716         alias = ll_splice_alias(inode, *de);
717         if (IS_ERR(alias))
718                 GOTO(out, rc = PTR_ERR(alias));
719
720         *de = alias;
721
722         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
723                 /* we have lookup look - unhide dentry */
724                 if (bits & MDS_INODELOCK_LOOKUP)
725                         d_lustre_revalidate(*de);
726
727                 if (encrypt) {
728                         rc = llcrypt_get_encryption_info(inode);
729                         if (rc)
730                                 GOTO(out, rc);
731                         if (!llcrypt_has_encryption_key(inode))
732                                 GOTO(out, rc = -ENOKEY);
733                 }
734         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
735                 /*
736                  * If file was created on the server, the dentry is revalidated
737                  * in ll_create_it if the lock allows for it.
738                  */
739                 /* Check that parent has UPDATE lock. */
740                 struct lookup_intent parent_it = {
741                                         .it_op = IT_GETATTR,
742                                         .it_lock_handle = 0 };
743                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
744
745                 /* If it is striped directory, get the real stripe parent */
746                 if (unlikely(ll_dir_striped(parent))) {
747                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
748                                                  ll_i2info(parent)->lli_lsm_md,
749                                                  (*de)->d_name.name,
750                                                  (*de)->d_name.len, &fid);
751                         if (rc != 0)
752                                 GOTO(out, rc);
753                 }
754
755                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
756                                        NULL)) {
757                         d_lustre_revalidate(*de);
758                         ll_intent_release(&parent_it);
759                 }
760         }
761
762         if (it_disposition(it, DISP_OPEN_CREATE)) {
763                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
764                                    ktime_us_delta(ktime_get(), kstart));
765         }
766
767         GOTO(out, rc = 0);
768
769 out:
770         if (rc != 0 && it->it_op & IT_OPEN) {
771                 ll_intent_drop_lock(it);
772                 ll_open_cleanup((*de)->d_sb, request);
773         }
774
775         return rc;
776 }
777
778 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
779                                    struct lookup_intent *it,
780                                    void **secctx, __u32 *secctxlen,
781                                    struct pcc_create_attach *pca,
782                                    bool encrypt,
783                                    void **encctx, __u32 *encctxlen)
784 {
785         ktime_t kstart = ktime_get();
786         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
787         struct dentry *save = dentry, *retval;
788         struct ptlrpc_request *req = NULL;
789         struct md_op_data *op_data = NULL;
790         struct lov_user_md *lum = NULL;
791         __u32 opc;
792         int rc;
793         char secctx_name[XATTR_NAME_MAX + 1];
794
795         ENTRY;
796
797         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
798                 RETURN(ERR_PTR(-ENAMETOOLONG));
799
800         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
801                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
802
803         if (d_mountpoint(dentry))
804                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
805
806         if (it == NULL || it->it_op == IT_GETXATTR)
807                 it = &lookup_it;
808
809         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
810                 rc = ll_revalidate_statahead(parent, &dentry, 0);
811                 if (rc == 1)
812                         RETURN(dentry == save ? NULL : dentry);
813         }
814
815         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
816             dentry->d_sb->s_flags & SB_RDONLY)
817                 RETURN(ERR_PTR(-EROFS));
818
819         if (it->it_op & IT_CREAT)
820                 opc = LUSTRE_OPC_CREATE;
821         else
822                 opc = LUSTRE_OPC_ANY;
823
824         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
825                                      dentry->d_name.len, 0, opc, NULL);
826         if (IS_ERR(op_data))
827                 GOTO(out, retval = ERR_CAST(op_data));
828
829         /* enforce umask if acl disabled or MDS doesn't support umask */
830         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
831                 it->it_create_mode &= ~current_umask();
832
833         if (it->it_op & IT_CREAT &&
834             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
835                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
836                                              &dentry->d_name,
837                                              &op_data->op_file_secctx_name,
838                                              &op_data->op_file_secctx,
839                                              &op_data->op_file_secctx_size);
840                 if (rc < 0)
841                         GOTO(out, retval = ERR_PTR(rc));
842                 if (secctx != NULL)
843                         *secctx = op_data->op_file_secctx;
844                 if (secctxlen != NULL)
845                         *secctxlen = op_data->op_file_secctx_size;
846         } else {
847                 if (secctx != NULL)
848                         *secctx = NULL;
849                 if (secctxlen != NULL)
850                         *secctxlen = 0;
851         }
852         if (it->it_op & IT_CREAT && encrypt) {
853                 rc = llcrypt_inherit_context(parent, NULL, op_data, false);
854                 if (rc)
855                         GOTO(out, retval = ERR_PTR(rc));
856                 if (encctx != NULL)
857                         *encctx = op_data->op_file_encctx;
858                 if (encctxlen != NULL)
859                         *encctxlen = op_data->op_file_encctx_size;
860         } else {
861                 if (encctx != NULL)
862                         *encctx = NULL;
863                 if (encctxlen != NULL)
864                         *encctxlen = 0;
865         }
866
867         /* ask for security context upon intent */
868         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
869                 /* get name of security xattr to request to server */
870                 rc = ll_listsecurity(parent, secctx_name,
871                                      sizeof(secctx_name));
872                 if (rc < 0) {
873                         CDEBUG(D_SEC, "cannot get security xattr name for "
874                                DFID": rc = %d\n",
875                                PFID(ll_inode2fid(parent)), rc);
876                 } else if (rc > 0) {
877                         op_data->op_file_secctx_name = secctx_name;
878                         op_data->op_file_secctx_name_size = rc;
879                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
880                                rc, secctx_name, PFID(ll_inode2fid(parent)));
881                 }
882         }
883
884         if (pca && pca->pca_dataset) {
885                 struct pcc_dataset *dataset = pca->pca_dataset;
886
887                 OBD_ALLOC_PTR(lum);
888                 if (lum == NULL)
889                         GOTO(out, retval = ERR_PTR(-ENOMEM));
890
891                 lum->lmm_magic = LOV_USER_MAGIC_V1;
892                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
893                 op_data->op_data = lum;
894                 op_data->op_data_size = sizeof(*lum);
895                 op_data->op_archive_id = dataset->pccd_rwid;
896
897                 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
898                                    op_data);
899                 if (rc)
900                         GOTO(out, retval = ERR_PTR(rc));
901
902                 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
903                                       &pca->pca_dentry);
904                 if (rc)
905                         GOTO(out, retval = ERR_PTR(rc));
906
907                 it->it_flags |= MDS_OPEN_PCC;
908         }
909
910         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
911                             &ll_md_blocking_ast, 0);
912         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
913          * client does not know which suppgid should be sent to the MDS, or
914          * some other(s) changed the target file's GID after this RPC sent
915          * to the MDS with the suppgid as the original GID, then we should
916          * try again with right suppgid. */
917         if (rc == -EACCES && it->it_op & IT_OPEN &&
918             it_disposition(it, DISP_OPEN_DENY)) {
919                 struct mdt_body *body;
920
921                 LASSERT(req != NULL);
922
923                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
924                 if (op_data->op_suppgids[0] == body->mbo_gid ||
925                     op_data->op_suppgids[1] == body->mbo_gid ||
926                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
927                         GOTO(out, retval = ERR_PTR(-EACCES));
928
929                 fid_zero(&op_data->op_fid2);
930                 op_data->op_suppgids[1] = body->mbo_gid;
931                 ptlrpc_req_finished(req);
932                 req = NULL;
933                 ll_intent_release(it);
934                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
935                                     &ll_md_blocking_ast, 0);
936         }
937
938         if (rc < 0)
939                 GOTO(out, retval = ERR_PTR(rc));
940
941         /* dir layout may change */
942         ll_unlock_md_op_lsm(op_data);
943         rc = ll_lookup_it_finish(req, it, parent, &dentry,
944                                  secctx != NULL ? *secctx : NULL,
945                                  secctxlen != NULL ? *secctxlen : 0,
946                                  encctx != NULL ? *encctx : NULL,
947                                  encctxlen != NULL ? *encctxlen : 0,
948                                  kstart, encrypt);
949         if (rc != 0) {
950                 ll_intent_release(it);
951                 GOTO(out, retval = ERR_PTR(rc));
952         }
953
954         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
955             !S_ISREG(dentry->d_inode->i_mode) &&
956             !S_ISDIR(dentry->d_inode->i_mode)) {
957                 ll_release_openhandle(dentry, it);
958         }
959         ll_lookup_finish_locks(it, dentry);
960
961         GOTO(out, retval = (dentry == save) ? NULL : dentry);
962
963 out:
964         if (op_data != NULL && !IS_ERR(op_data)) {
965                 if (secctx != NULL && secctxlen != NULL) {
966                         /* caller needs sec ctx info, so reset it in op_data to
967                          * prevent it from being freed */
968                         op_data->op_file_secctx = NULL;
969                         op_data->op_file_secctx_size = 0;
970                 }
971                 if (encctx != NULL && encctxlen != NULL &&
972                     it->it_op & IT_CREAT && encrypt) {
973                         /* caller needs enc ctx info, so reset it in op_data to
974                          * prevent it from being freed
975                          */
976                         op_data->op_file_encctx = NULL;
977                         op_data->op_file_encctx_size = 0;
978                 }
979                 ll_finish_md_op_data(op_data);
980         }
981
982         if (lum != NULL)
983                 OBD_FREE_PTR(lum);
984
985         ptlrpc_req_finished(req);
986         return retval;
987 }
988
989 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
990                                    unsigned int flags)
991 {
992         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
993         struct dentry *de;
994
995         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
996                dentry, PFID(ll_inode2fid(parent)), parent, flags);
997
998         /*
999          * Optimize away (CREATE && !OPEN). Let .create handle the race.
1000          * but only if we have write permissions there, otherwise we need
1001          * to proceed with lookup. LU-4185
1002          */
1003         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1004             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
1005                 return NULL;
1006
1007         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1008                 itp = NULL;
1009         else
1010                 itp = &it;
1011         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1012                           NULL, NULL);
1013
1014         if (itp != NULL)
1015                 ll_intent_release(itp);
1016
1017         return de;
1018 }
1019
1020 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1021 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
1022 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
1023 # define ll_last_arg
1024 # define ll_set_created(o, f)                                           \
1025 do {                                                                    \
1026         (f)->f_mode |= FMODE_CREATED;                                   \
1027 } while (0)
1028
1029 #else
1030 # define ll_is_opened(o, f)             (*(o))
1031 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
1032 # define ll_last_arg                    , int *opened
1033 # define ll_set_created(o, f)                                           \
1034 do {                                                                    \
1035         *(o) |= FILE_CREATED;                                           \
1036 } while (0)
1037
1038 #endif
1039
1040 /*
1041  * For cached negative dentry and new dentry, handle lookup/create/open
1042  * together.
1043  */
1044 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1045                           struct file *file, unsigned open_flags,
1046                           umode_t mode ll_last_arg)
1047 {
1048         struct lookup_intent *it;
1049         struct dentry *de;
1050         long long lookup_flags = LOOKUP_OPEN;
1051         void *secctx = NULL;
1052         __u32 secctxlen = 0;
1053         void *encctx = NULL;
1054         __u32 encctxlen = 0;
1055         struct ll_sb_info *sbi = NULL;
1056         struct pcc_create_attach pca = { NULL, NULL };
1057         bool encrypt = false;
1058         int rc = 0;
1059         ENTRY;
1060
1061         CDEBUG(D_VFSTRACE,
1062                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1063                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1064                ll_is_opened(opened, file));
1065
1066         /* Only negative dentries enter here */
1067         LASSERT(dentry->d_inode == NULL);
1068
1069         if (!d_unhashed(dentry)) {
1070                 /* A valid negative dentry that just passed revalidation,
1071                  * there's little point to try and open it server-side,
1072                  * even though there's a minuscule chance it might succeed.
1073                  * Either way it's a valid race to just return -ENOENT here.
1074                  */
1075                 if (!(open_flags & O_CREAT))
1076                         return -ENOENT;
1077
1078                 /* Otherwise we just unhash it to be rehashed afresh via
1079                  * lookup if necessary
1080                  */
1081                 d_drop(dentry);
1082         }
1083
1084         OBD_ALLOC(it, sizeof(*it));
1085         if (!it)
1086                 RETURN(-ENOMEM);
1087
1088         it->it_op = IT_OPEN;
1089         if (open_flags & O_CREAT) {
1090                 it->it_op |= IT_CREAT;
1091                 lookup_flags |= LOOKUP_CREATE;
1092                 sbi = ll_i2sbi(dir);
1093                 /* Volatile file is used for HSM restore, so do not use PCC */
1094                 if (!filename_is_volatile(dentry->d_name.name,
1095                                           dentry->d_name.len, NULL)) {
1096                         struct pcc_matcher item;
1097                         struct pcc_dataset *dataset;
1098
1099                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1100                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1101                         item.pm_projid = ll_i2info(dir)->lli_projid;
1102                         item.pm_name = &dentry->d_name;
1103                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1104                                                         &item);
1105                         pca.pca_dataset = dataset;
1106                 }
1107         }
1108         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1109         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1110         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1111
1112         if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1113                 /* we know that we are going to create a regular file because
1114                  * we set S_IFREG bit on it->it_create_mode above
1115                  */
1116                 rc = llcrypt_get_encryption_info(dir);
1117                 if (rc)
1118                         GOTO(out_release, rc);
1119                 if (!llcrypt_has_encryption_key(dir))
1120                         GOTO(out_release, rc = -ENOKEY);
1121                 encrypt = true;
1122                 rc = 0;
1123         }
1124
1125         /* Dentry added to dcache tree in ll_lookup_it */
1126         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1127                           &encctx, &encctxlen);
1128         if (IS_ERR(de))
1129                 rc = PTR_ERR(de);
1130         else if (de != NULL)
1131                 dentry = de;
1132
1133         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1134
1135         if (!rc) {
1136                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1137                         /* Dentry instantiated in ll_create_it. */
1138                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1139                                           encrypt, encctx, encctxlen);
1140                         security_release_secctx(secctx, secctxlen);
1141                         llcrypt_free_ctx(encctx, encctxlen);
1142                         if (rc) {
1143                                 /* We dget in ll_splice_alias. */
1144                                 if (de != NULL)
1145                                         dput(de);
1146                                 goto out_release;
1147                         }
1148
1149                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1150                         if (rc) {
1151                                 if (de != NULL)
1152                                         dput(de);
1153                                 GOTO(out_release, rc);
1154                         }
1155
1156                         ll_set_created(opened, file);
1157                 } else {
1158                         /* Open the file with O_CREAT, but the file already
1159                          * existed on MDT. This may happend in the case that
1160                          * the LOOKUP ibits lock is revoked and the
1161                          * corresponding dentry cache is deleted.
1162                          * i.e. In the current Lustre, the truncate operation
1163                          * will revoke the LOOKUP ibits lock, and the file
1164                          * dentry cache will be invalidated. The following open
1165                          * with O_CREAT flag will call into ->atomic_open, the
1166                          * file was wrongly though as newly created file and
1167                          * try to auto cache the file. So after client knows it
1168                          * is not a DISP_OPEN_CREATE, it should cleanup the
1169                          * already created PCC copy.
1170                          */
1171                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1172
1173                         if (open_flags & O_CREAT && encrypt &&
1174                             dentry->d_inode) {
1175                                 rc = ll_set_encflags(dentry->d_inode, encctx,
1176                                                      encctxlen, true);
1177                                 llcrypt_free_ctx(encctx, encctxlen);
1178                                 if (rc)
1179                                         GOTO(out_release, rc);
1180                         }
1181                 }
1182
1183                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1184                         /* Open dentry. */
1185                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1186                                 /* We cannot call open here as it might
1187                                  * deadlock. This case is unreachable in
1188                                  * practice because of OBD_CONNECT_NODEVOH. */
1189                                 rc = finish_no_open(file, de);
1190                         } else {
1191                                 file->private_data = it;
1192                                 rc = ll_finish_open(file, dentry, opened);
1193                                 /* We dget in ll_splice_alias. finish_open takes
1194                                  * care of dget for fd open.
1195                                  */
1196                                 if (de != NULL)
1197                                         dput(de);
1198                         }
1199                 } else {
1200                         rc = finish_no_open(file, de);
1201                 }
1202         } else {
1203                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1204         }
1205
1206 out_release:
1207         ll_intent_release(it);
1208         OBD_FREE(it, sizeof(*it));
1209
1210         RETURN(rc);
1211 }
1212
1213 /* We depend on "mode" being set with the proper file type/umask by now */
1214 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1215 {
1216         struct inode *inode = NULL;
1217         struct ptlrpc_request *request = NULL;
1218         struct ll_sb_info *sbi = ll_i2sbi(dir);
1219         int rc;
1220         ENTRY;
1221
1222         LASSERT(it && it->it_disposition);
1223
1224         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1225         request = it->it_request;
1226         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1227         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1228         if (rc)
1229                 GOTO(out, inode = ERR_PTR(rc));
1230
1231         /* Pause to allow for a race with concurrent access by fid */
1232         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1233
1234         /* We asked for a lock on the directory, but were granted a
1235          * lock on the inode.  Since we finally have an inode pointer,
1236          * stuff it in the lock. */
1237         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1238                PFID(ll_inode2fid(inode)), inode);
1239         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1240         EXIT;
1241  out:
1242         ptlrpc_req_finished(request);
1243         return inode;
1244 }
1245
1246 /*
1247  * By the time this is called, we already have created the directory cache
1248  * entry for the new file, but it is so far negative - it has no inode.
1249  *
1250  * We defer creating the OBD object(s) until open, to keep the intent and
1251  * non-intent code paths similar, and also because we do not have the MDS
1252  * inode number before calling ll_create_node() (which is needed for LOV),
1253  * so we would need to do yet another RPC to the MDS to store the LOV EA
1254  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1255  * lmm_size in datalen (the MDS still has code which will handle that).
1256  *
1257  * If the create succeeds, we fill in the inode information
1258  * with d_instantiate().
1259  */
1260 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1261                         struct lookup_intent *it,
1262                         void *secctx, __u32 secctxlen, bool encrypt,
1263                         void *encctx, __u32 encctxlen)
1264 {
1265         struct inode *inode;
1266         __u64 bits = 0;
1267         int rc = 0;
1268         ENTRY;
1269
1270         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1271                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1272
1273         rc = it_open_error(DISP_OPEN_CREATE, it);
1274         if (rc)
1275                 RETURN(rc);
1276
1277         inode = ll_create_node(dir, it);
1278         if (IS_ERR(inode))
1279                 RETURN(PTR_ERR(inode));
1280
1281         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1282             secctx != NULL) {
1283                 /* must be done before d_instantiate, because it calls
1284                  * security_d_instantiate, which means a getxattr if security
1285                  * context is not set yet */
1286                 /* no need to protect selinux_inode_setsecurity() by
1287                  * inode_lock. Taking it would lead to a client deadlock
1288                  * LU-13617
1289                  */
1290                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1291                 if (rc)
1292                         RETURN(rc);
1293         }
1294
1295         d_instantiate(dentry, inode);
1296
1297         if (encrypt) {
1298                 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1299                 if (rc)
1300                         RETURN(rc);
1301         }
1302
1303         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1304                 rc = ll_inode_init_security(dentry, inode, dir);
1305                 if (rc)
1306                         RETURN(rc);
1307         }
1308
1309         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1310         if (bits & MDS_INODELOCK_LOOKUP)
1311                 d_lustre_revalidate(dentry);
1312
1313         RETURN(0);
1314 }
1315
1316 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1317 {
1318         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1319                                                        &RMF_MDT_BODY);
1320
1321         LASSERT(body);
1322         if (body->mbo_valid & OBD_MD_FLMTIME &&
1323             body->mbo_mtime > inode->i_mtime.tv_sec) {
1324                 CDEBUG(D_INODE,
1325                        "setting fid " DFID " mtime from %lld to %llu\n",
1326                        PFID(ll_inode2fid(inode)),
1327                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1328                 inode->i_mtime.tv_sec = body->mbo_mtime;
1329         }
1330
1331         if (body->mbo_valid & OBD_MD_FLCTIME &&
1332             body->mbo_ctime > inode->i_ctime.tv_sec)
1333                 inode->i_ctime.tv_sec = body->mbo_ctime;
1334 }
1335
1336 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1337                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1338 {
1339         struct qstr *name = &dchild->d_name;
1340         struct ptlrpc_request *request = NULL;
1341         struct md_op_data *op_data = NULL;
1342         struct inode *inode = NULL;
1343         struct ll_sb_info *sbi = ll_i2sbi(dir);
1344         int tgt_len = 0;
1345         bool encrypt = false;
1346         int err;
1347
1348         ENTRY;
1349         if (unlikely(tgt != NULL))
1350                 tgt_len = strlen(tgt) + 1;
1351
1352 again:
1353         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1354                                      name->len, 0, opc, NULL);
1355         if (IS_ERR(op_data))
1356                 GOTO(err_exit, err = PTR_ERR(op_data));
1357
1358         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1359                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1360                                               &op_data->op_file_secctx_name,
1361                                               &op_data->op_file_secctx,
1362                                               &op_data->op_file_secctx_size);
1363                 if (err < 0)
1364                         GOTO(err_exit, err);
1365         }
1366
1367         if (ll_sbi_has_encrypt(sbi) &&
1368             ((IS_ENCRYPTED(dir) &&
1369             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1370             (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1371                 err = llcrypt_get_encryption_info(dir);
1372                 if (err)
1373                         GOTO(err_exit, err);
1374                 if (!llcrypt_has_encryption_key(dir))
1375                         GOTO(err_exit, err = -ENOKEY);
1376                 encrypt = true;
1377         }
1378
1379         if (encrypt) {
1380                 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1381                 if (err)
1382                         GOTO(err_exit, err);
1383         }
1384
1385         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1386                         from_kuid(&init_user_ns, current_fsuid()),
1387                         from_kgid(&init_user_ns, current_fsgid()),
1388                         cfs_curproc_cap_pack(), rdev, &request);
1389 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1390         /*
1391          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1392          * fetch default LMV here.
1393          */
1394         if (unlikely(err == -EREMOTE)) {
1395                 struct ll_inode_info    *lli = ll_i2info(dir);
1396                 struct lmv_user_md      *lum;
1397                 int                     lumsize;
1398                 int                     err2;
1399
1400                 ptlrpc_req_finished(request);
1401                 request = NULL;
1402
1403                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1404                                         OBD_MD_DEFAULT_MEA);
1405                 ll_finish_md_op_data(op_data);
1406                 op_data = NULL;
1407                 if (err2 == 0) {
1408                         struct lustre_md md = { NULL };
1409
1410                         md.body = req_capsule_server_get(&request->rq_pill,
1411                                                          &RMF_MDT_BODY);
1412                         if (!md.body)
1413                                 GOTO(err_exit, err = -EPROTO);
1414
1415                         OBD_ALLOC_PTR(md.default_lmv);
1416                         if (!md.default_lmv)
1417                                 GOTO(err_exit, err = -ENOMEM);
1418
1419                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1420                         md.default_lmv->lsm_md_stripe_count =
1421                                 lum->lum_stripe_count;
1422                         md.default_lmv->lsm_md_master_mdt_index =
1423                                 lum->lum_stripe_offset;
1424                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1425
1426                         err = ll_update_inode(dir, &md);
1427                         md_free_lustre_md(sbi->ll_md_exp, &md);
1428                         if (err)
1429                                 GOTO(err_exit, err);
1430                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1431                         /*
1432                          * If there are no default stripe EA on the MDT, but the
1433                          * client has default stripe, then it probably means
1434                          * default stripe EA has just been deleted.
1435                          */
1436                         down_write(&lli->lli_lsm_sem);
1437                         if (lli->lli_default_lsm_md)
1438                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1439                         lli->lli_default_lsm_md = NULL;
1440                         up_write(&lli->lli_lsm_sem);
1441                 } else {
1442                         GOTO(err_exit, err);
1443                 }
1444
1445                 ptlrpc_req_finished(request);
1446                 request = NULL;
1447                 goto again;
1448         }
1449 #endif
1450
1451         if (err < 0)
1452                 GOTO(err_exit, err);
1453
1454         ll_update_times(request, dir);
1455
1456         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1457
1458         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1459         if (err)
1460                 GOTO(err_exit, err);
1461
1462         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1463                 /* must be done before d_instantiate, because it calls
1464                  * security_d_instantiate, which means a getxattr if security
1465                  * context is not set yet */
1466                 /* no need to protect selinux_inode_setsecurity() by
1467                  * inode_lock. Taking it would lead to a client deadlock
1468                  * LU-13617
1469                  */
1470                 err = security_inode_notifysecctx(inode,
1471                                                   op_data->op_file_secctx,
1472                                                   op_data->op_file_secctx_size);
1473                 if (err)
1474                         GOTO(err_exit, err);
1475         }
1476
1477         d_instantiate(dchild, inode);
1478
1479         if (encrypt) {
1480                 err = ll_set_encflags(inode, op_data->op_file_encctx,
1481                                       op_data->op_file_encctx_size, true);
1482                 if (err)
1483                         GOTO(err_exit, err);
1484         }
1485
1486         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1487                 err = ll_inode_init_security(dchild, inode, dir);
1488                 if (err)
1489                         GOTO(err_exit, err);
1490         }
1491
1492         EXIT;
1493 err_exit:
1494         if (request != NULL)
1495                 ptlrpc_req_finished(request);
1496
1497         if (!IS_ERR_OR_NULL(op_data))
1498                 ll_finish_md_op_data(op_data);
1499
1500         RETURN(err);
1501 }
1502
1503 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1504                     dev_t rdev)
1505 {
1506         ktime_t kstart = ktime_get();
1507         int err;
1508         ENTRY;
1509
1510         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1511                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1512
1513         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1514                 mode &= ~current_umask();
1515
1516         switch (mode & S_IFMT) {
1517         case 0:
1518                 mode |= S_IFREG;
1519                 /* fallthrough */
1520         case S_IFREG:
1521         case S_IFCHR:
1522         case S_IFBLK:
1523         case S_IFIFO:
1524         case S_IFSOCK:
1525                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1526                                   LUSTRE_OPC_MKNOD);
1527                 break;
1528         case S_IFDIR:
1529                 err = -EPERM;
1530                 break;
1531         default:
1532                 err = -EINVAL;
1533         }
1534
1535         if (!err)
1536                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1537                                    ktime_us_delta(ktime_get(), kstart));
1538
1539         RETURN(err);
1540 }
1541
1542 /*
1543  * Plain create. Intent create is handled in atomic_open.
1544  */
1545 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1546                         umode_t mode, bool want_excl)
1547 {
1548         ktime_t kstart = ktime_get();
1549         int rc;
1550
1551         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1552
1553         CDEBUG(D_VFSTRACE,
1554                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1555                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1556
1557         /* Using mknod(2) to create a regular file is designed to not recognize
1558          * volatile file name, so we use ll_mknod() here. */
1559         rc = ll_mknod(dir, dentry, mode, 0);
1560
1561         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1562                dentry, d_unhashed(dentry));
1563
1564         if (!rc)
1565                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1566                                    ktime_us_delta(ktime_get(), kstart));
1567
1568         return rc;
1569 }
1570
1571 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1572                       const char *oldpath)
1573 {
1574         ktime_t kstart = ktime_get();
1575         int err;
1576         ENTRY;
1577
1578         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1579                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1580
1581         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1582                           LUSTRE_OPC_SYMLINK);
1583
1584         if (!err)
1585                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1586                                    ktime_us_delta(ktime_get(), kstart));
1587
1588         RETURN(err);
1589 }
1590
1591 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1592                    struct dentry *new_dentry)
1593 {
1594         struct inode *src = old_dentry->d_inode;
1595         struct qstr *name = &new_dentry->d_name;
1596         struct ll_sb_info *sbi = ll_i2sbi(dir);
1597         struct ptlrpc_request *request = NULL;
1598         struct md_op_data *op_data;
1599         ktime_t kstart = ktime_get();
1600         int err;
1601
1602         ENTRY;
1603         CDEBUG(D_VFSTRACE,
1604                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1605                PFID(ll_inode2fid(src)), src,
1606                PFID(ll_inode2fid(dir)), dir, new_dentry);
1607
1608         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1609         if (err)
1610                 RETURN(err);
1611
1612         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1613                                      0, LUSTRE_OPC_ANY, NULL);
1614         if (IS_ERR(op_data))
1615                 RETURN(PTR_ERR(op_data));
1616
1617         err = md_link(sbi->ll_md_exp, op_data, &request);
1618         ll_finish_md_op_data(op_data);
1619         if (err)
1620                 GOTO(out, err);
1621
1622         ll_update_times(request, dir);
1623         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1624                            ktime_us_delta(ktime_get(), kstart));
1625         EXIT;
1626 out:
1627         ptlrpc_req_finished(request);
1628         RETURN(err);
1629 }
1630
1631 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1632 {
1633         ktime_t kstart = ktime_get();
1634         int err;
1635         ENTRY;
1636
1637         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1638                dchild, PFID(ll_inode2fid(dir)), dir);
1639
1640         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1641                 mode &= ~current_umask();
1642
1643         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1644
1645         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1646         if (err == 0)
1647                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1648                                    ktime_us_delta(ktime_get(), kstart));
1649
1650         RETURN(err);
1651 }
1652
1653 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1654 {
1655         struct qstr *name = &dchild->d_name;
1656         struct ptlrpc_request *request = NULL;
1657         struct md_op_data *op_data;
1658         ktime_t kstart = ktime_get();
1659         int rc;
1660
1661         ENTRY;
1662
1663         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1664                dchild, PFID(ll_inode2fid(dir)), dir);
1665
1666         if (unlikely(d_mountpoint(dchild)))
1667                 RETURN(-EBUSY);
1668
1669         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1670                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1671         if (IS_ERR(op_data))
1672                 RETURN(PTR_ERR(op_data));
1673
1674         if (dchild->d_inode != NULL)
1675                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1676
1677         op_data->op_fid2 = op_data->op_fid3;
1678         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1679         ll_finish_md_op_data(op_data);
1680         if (!rc)
1681                 ll_update_times(request, dir);
1682
1683         ptlrpc_req_finished(request);
1684         if (!rc)
1685                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1686                                    ktime_us_delta(ktime_get(), kstart));
1687         RETURN(rc);
1688 }
1689
1690 /**
1691  * Remove dir entry
1692  **/
1693 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1694 {
1695         struct ptlrpc_request *request = NULL;
1696         struct md_op_data *op_data;
1697         ktime_t kstart = ktime_get();
1698         int rc;
1699         ENTRY;
1700
1701         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1702                namelen, name, PFID(ll_inode2fid(dir)), dir);
1703
1704         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1705                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1706         if (IS_ERR(op_data))
1707                 RETURN(PTR_ERR(op_data));
1708         op_data->op_cli_flags |= CLI_RM_ENTRY;
1709         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1710         ll_finish_md_op_data(op_data);
1711         if (!rc)
1712                 ll_update_times(request, dir);
1713
1714         ptlrpc_req_finished(request);
1715         if (!rc)
1716                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1717                                    ktime_us_delta(ktime_get(), kstart));
1718         RETURN(rc);
1719 }
1720
1721 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1722 {
1723         struct qstr *name = &dchild->d_name;
1724         struct ptlrpc_request *request = NULL;
1725         struct md_op_data *op_data;
1726         struct mdt_body *body;
1727         ktime_t kstart = ktime_get();
1728         int rc;
1729
1730         ENTRY;
1731
1732         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1733                dchild, PFID(ll_inode2fid(dir)), dir);
1734
1735         /*
1736          * XXX: unlink bind mountpoint maybe call to here,
1737          * just check it as vfs_unlink does.
1738          */
1739         if (unlikely(d_mountpoint(dchild)))
1740                 RETURN(-EBUSY);
1741
1742         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1743                                      LUSTRE_OPC_ANY, NULL);
1744         if (IS_ERR(op_data))
1745                 RETURN(PTR_ERR(op_data));
1746
1747         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1748         /* notify lower layer if inode has dirty pages */
1749         if (S_ISREG(dchild->d_inode->i_mode) &&
1750             ll_i2info(dchild->d_inode)->lli_clob &&
1751             dirty_cnt(dchild->d_inode))
1752                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1753         op_data->op_fid2 = op_data->op_fid3;
1754         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1755         ll_finish_md_op_data(op_data);
1756         if (rc)
1757                 GOTO(out, rc);
1758
1759         /*
1760          * The server puts attributes in on the last unlink, use them to update
1761          * the link count so the inode can be freed immediately.
1762          */
1763         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1764         if (body->mbo_valid & OBD_MD_FLNLINK)
1765                 set_nlink(dchild->d_inode, body->mbo_nlink);
1766
1767         ll_update_times(request, dir);
1768
1769 out:
1770         ptlrpc_req_finished(request);
1771         if (!rc)
1772                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1773                                    ktime_us_delta(ktime_get(), kstart));
1774         RETURN(rc);
1775 }
1776
1777 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1778                      struct inode *tgt, struct dentry *tgt_dchild
1779 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1780                      , unsigned int flags
1781 #endif
1782                      )
1783 {
1784         struct qstr *src_name = &src_dchild->d_name;
1785         struct qstr *tgt_name = &tgt_dchild->d_name;
1786         struct ptlrpc_request *request = NULL;
1787         struct ll_sb_info *sbi = ll_i2sbi(src);
1788         struct md_op_data *op_data;
1789         ktime_t kstart = ktime_get();
1790         int err;
1791         ENTRY;
1792
1793 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1794         if (flags)
1795                 return -EINVAL;
1796 #endif
1797
1798         CDEBUG(D_VFSTRACE,
1799                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1800                src_dchild, PFID(ll_inode2fid(src)), src,
1801                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1802
1803         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1804                 RETURN(-EBUSY);
1805
1806 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1807         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1808 #else
1809         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1810 #endif
1811         if (err)
1812                 RETURN(err);
1813
1814         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1815                                      LUSTRE_OPC_ANY, NULL);
1816         if (IS_ERR(op_data))
1817                 RETURN(PTR_ERR(op_data));
1818
1819         if (src_dchild->d_inode != NULL)
1820                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1821
1822         if (tgt_dchild->d_inode != NULL)
1823                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1824
1825         err = md_rename(sbi->ll_md_exp, op_data,
1826                         src_name->name, src_name->len,
1827                         tgt_name->name, tgt_name->len, &request);
1828         ll_finish_md_op_data(op_data);
1829         if (!err) {
1830                 ll_update_times(request, src);
1831                 ll_update_times(request, tgt);
1832         }
1833
1834         ptlrpc_req_finished(request);
1835
1836         if (!err) {
1837                 d_move(src_dchild, tgt_dchild);
1838                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1839                                    ktime_us_delta(ktime_get(), kstart));
1840         }
1841
1842         RETURN(err);
1843 }
1844
1845 const struct inode_operations ll_dir_inode_operations = {
1846         .mknod          = ll_mknod,
1847         .atomic_open    = ll_atomic_open,
1848         .lookup         = ll_lookup_nd,
1849         .create         = ll_create_nd,
1850         /* We need all these non-raw things for NFSD, to not patch it. */
1851         .unlink         = ll_unlink,
1852         .mkdir          = ll_mkdir,
1853         .rmdir          = ll_rmdir,
1854         .symlink        = ll_symlink,
1855         .link           = ll_link,
1856         .rename         = ll_rename,
1857         .setattr        = ll_setattr,
1858         .getattr        = ll_getattr,
1859         .permission     = ll_inode_permission,
1860 #ifdef HAVE_IOP_XATTR
1861         .setxattr       = ll_setxattr,
1862         .getxattr       = ll_getxattr,
1863         .removexattr    = ll_removexattr,
1864 #endif
1865         .listxattr      = ll_listxattr,
1866         .get_acl        = ll_get_acl,
1867 #ifdef HAVE_IOP_SET_ACL
1868         .set_acl        = ll_set_acl,
1869 #endif
1870 };
1871
1872 const struct inode_operations ll_special_inode_operations = {
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         .get_acl        = ll_get_acl,
1883 #ifdef HAVE_IOP_SET_ACL
1884         .set_acl        = ll_set_acl,
1885 #endif
1886 };