Whamcloud - gitweb
add support for build HEAD without vfs_intent patches.
[fs/lustre-release.git] / lustre / llite / namei.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  *  Copyright (c) 2002, 2003 Cluster File Systems, Inc.
5  *
6  *   This file is part of Lustre, http://www.lustre.org.
7  *
8  *   Lustre is free software; you can redistribute it and/or
9  *   modify it under the terms of version 2 of the GNU General Public
10  *   License as published by the Free Software Foundation.
11  *
12  *   Lustre is distributed in the hope that it will be useful,
13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  *   GNU General Public License for more details.
16  *
17  *   You should have received a copy of the GNU General Public License
18  *   along with Lustre; if not, write to the Free Software
19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20  */
21
22 #include <linux/fs.h>
23 #include <linux/sched.h>
24 #include <linux/mm.h>
25 #include <linux/smp_lock.h>
26 #include <linux/quotaops.h>
27 #include <linux/highmem.h>
28 #include <linux/pagemap.h>
29
30 #define DEBUG_SUBSYSTEM S_LLITE
31
32 #include <obd_support.h>
33 #include <lustre_fid.h>
34 #include <lustre_lite.h>
35 #include <lustre_dlm.h>
36 #include <lustre_ver.h>
37 #include <lustre_mdc.h>
38 #include "llite_internal.h"
39
40 /* methods */
41 extern struct dentry_operations ll_d_ops;
42
43 /*
44  * Check if we have something mounted at the named dchild.
45  * In such a case there would always be dentry present.
46  */
47 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
48                            struct qstr *name)
49 {
50         int mounted = 0;
51
52         if (unlikely(dchild)) {
53                 mounted = d_mountpoint(dchild);
54         } else if (dparent) {
55                 dchild = d_lookup(dparent, name);
56                 if (dchild) {
57                         mounted = d_mountpoint(dchild);
58                         dput(dchild);
59                 }
60         }
61         return mounted;
62 }
63
64 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
65 {
66         ENTRY;
67
68         ldlm_lock_decref(lockh, mode);
69
70         RETURN(0);
71 }
72
73 /*
74  * Get an inode by inode number (already instantiated by the intent lookup).
75  * Returns inode or NULL
76  */
77 struct inode *ll_iget(struct super_block *sb, ino_t hash,
78                       struct lustre_md *md)
79 {
80         struct ll_inode_info *lli;
81         struct inode *inode;
82         LASSERT(hash != 0);
83
84         inode = iget_locked(sb, hash);
85         if (inode) {
86                 if (inode->i_state & I_NEW) {
87                         lli = ll_i2info(inode);
88                         ll_read_inode2(inode, md);
89                         unlock_new_inode(inode);
90                 } else {
91                         if (!(inode->i_state & (I_FREEING | I_CLEAR)))
92                                 ll_update_inode(inode, md);
93                 }
94                 CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n",
95                        inode->i_ino, inode->i_generation, inode);
96         }
97
98         return inode;
99 }
100
101 static void ll_drop_negative_dentry(struct inode *dir)
102
103         struct dentry *dentry, *tmp_alias, *tmp_subdir;
104
105         spin_lock(&dcache_lock);
106 restart:
107         list_for_each_entry_safe(dentry, tmp_alias,
108                                  &dir->i_dentry,d_alias) {
109                 if (!list_empty(&dentry->d_subdirs)) {
110                         struct dentry *child;
111                         list_for_each_entry_safe(child, tmp_subdir,
112                                                  &dentry->d_subdirs,
113                                                  d_child) {
114                                 /* XXX Print some debug here? */
115                                 if (!child->d_inode)
116                                 /* Negative dentry. If we were
117                                    dropping dcache lock, go
118                                    throught the list again */
119                                         if (ll_drop_dentry(child))
120                                                 goto restart;
121                         }
122                 }
123         }
124         spin_unlock(&dcache_lock);
125 }
126
127
128 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
129                        void *data, int flag)
130 {
131         int rc;
132         struct lustre_handle lockh;
133         ENTRY;
134
135         switch (flag) {
136         case LDLM_CB_BLOCKING:
137                 ldlm_lock2handle(lock, &lockh);
138                 rc = ldlm_cli_cancel(&lockh);
139                 if (rc < 0) {
140                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
141                         RETURN(rc);
142                 }
143                 break;
144         case LDLM_CB_CANCELING: {
145                 struct inode *inode = ll_inode_from_lock(lock);
146                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
147                 struct lu_fid *fid;
148
149                 /* Invalidate all dentries associated with this inode */
150                 if (inode == NULL)
151                         break;
152
153                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
154                 if ((bits & MDS_INODELOCK_LOOKUP) &&
155                     ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
156                         bits &= ~MDS_INODELOCK_LOOKUP;
157                 if ((bits & MDS_INODELOCK_UPDATE) &&
158                     ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
159                         bits &= ~MDS_INODELOCK_UPDATE;
160                 if ((bits & MDS_INODELOCK_OPEN) &&
161                     ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
162                         bits &= ~MDS_INODELOCK_OPEN;
163
164                 fid = ll_inode2fid(inode);
165                 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
166                     lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
167                     lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
168                         LDLM_ERROR(lock, "data mismatch with object "
169                                    DFID" (%p)", PFID(fid), inode);
170                 }
171
172                 if (bits & MDS_INODELOCK_OPEN) {
173                         int flags = 0;
174                         switch (lock->l_req_mode) {
175                         case LCK_CW:
176                                 flags = FMODE_WRITE;
177                                 break;
178                         case LCK_PR:
179                                 flags = FMODE_EXEC;
180                                 break;
181                         case LCK_CR:
182                                 flags = FMODE_READ;
183                                 break;
184                         default:
185                                 CERROR("Unexpected lock mode for OPEN lock "
186                                        "%d, inode %ld\n", lock->l_req_mode,
187                                        inode->i_ino);
188                         }
189                         ll_md_real_close(inode, flags);
190                 }
191
192                 if (bits & MDS_INODELOCK_UPDATE)
193                         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
194
195                 if (S_ISDIR(inode->i_mode) &&
196                      (bits & MDS_INODELOCK_UPDATE)) {
197                         CDEBUG(D_INODE, "invalidating inode %lu\n",
198                                inode->i_ino);
199                         truncate_inode_pages(inode->i_mapping, 0);
200                         ll_drop_negative_dentry(inode);
201                 }
202
203                 if (inode->i_sb->s_root &&
204                     inode != inode->i_sb->s_root->d_inode &&
205                     (bits & MDS_INODELOCK_LOOKUP))
206                         ll_unhash_aliases(inode);
207                 iput(inode);
208                 break;
209         }
210         default:
211                 LBUG();
212         }
213
214         RETURN(0);
215 }
216
217 /* Pack the required supplementary groups into the supplied groups array.
218  * If we don't need to use the groups from the target inode(s) then we
219  * instead pack one or more groups from the user's supplementary group
220  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
221 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
222 {
223         int i;
224
225         LASSERT(i1 != NULL);
226         LASSERT(suppgids != NULL);
227
228         if (in_group_p(i1->i_gid))
229                 suppgids[0] = i1->i_gid;
230         else
231                 suppgids[0] = -1;
232
233         if (i2) {
234                 if (in_group_p(i2->i_gid))
235                         suppgids[1] = i2->i_gid;
236                 else
237                         suppgids[1] = -1;
238         } else {
239                 suppgids[1] = -1;
240         }
241
242         for (i = 0; i < current_ngroups; i++) {
243                 if (suppgids[0] == -1) {
244                         if (current_groups[i] != suppgids[1])
245                                 suppgids[0] = current_groups[i];
246                         continue;
247                 }
248                 if (suppgids[1] == -1) {
249                         if (current_groups[i] != suppgids[0])
250                                 suppgids[1] = current_groups[i];
251                         continue;
252                 }
253                 break;
254         }
255 }
256
257 static void ll_d_add(struct dentry *de, struct inode *inode)
258 {
259         CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
260         /* d_instantiate */
261         if (!list_empty(&de->d_alias)) {
262                 spin_unlock(&dcache_lock);
263                 CERROR("dentry %.*s %p alias next %p, prev %p\n",
264                        de->d_name.len, de->d_name.name, de,
265                        de->d_alias.next, de->d_alias.prev);
266                 LBUG();
267         }
268         if (inode)
269                 list_add(&de->d_alias, &inode->i_dentry);
270         de->d_inode = inode;
271
272         /* d_rehash */
273         if (!d_unhashed(de)) {
274                 spin_unlock(&dcache_lock);
275                 CERROR("dentry %.*s %p hash next %p\n",
276                        de->d_name.len, de->d_name.name, de, de->d_hash.next);
277                 LBUG();
278         }
279         __d_rehash(de, 0);
280 }
281
282 /* Search "inode"'s alias list for a dentry that has the same name and parent
283  * as de.  If found, return it.  If not found, return de.
284  * Lustre can't use d_add_unique because don't unhash aliases for directory
285  * in ll_revalidate_it.  After revaliadate inode will be have hashed aliases
286  * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
287  */
288 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
289 {
290         struct list_head *tmp;
291         struct dentry *dentry;
292         struct dentry *last_discon = NULL;
293  
294         spin_lock(&dcache_lock);
295         list_for_each(tmp, &inode->i_dentry) {
296                 dentry = list_entry(tmp, struct dentry, d_alias);
297
298                 /* We are called here with 'de' already on the aliases list. */
299                 if (unlikely(dentry == de)) {
300                         CERROR("whoops\n");
301                         continue;
302                 }
303
304                 if (dentry->d_flags & DCACHE_DISCONNECTED) {
305                         LASSERT(last_discon == NULL);
306                         last_discon = dentry;
307                         continue;
308                 }
309
310                 if (dentry->d_parent != de->d_parent)
311                         continue;
312
313                 if (dentry->d_name.hash != de->d_name.hash)
314                         continue;
315
316                 if (dentry->d_name.len != de->d_name.len)
317                         continue;
318
319                 if (memcmp(dentry->d_name.name, de->d_name.name,
320                            de->d_name.len) != 0)
321                         continue;
322
323                 dget_locked(dentry);
324                 lock_dentry(dentry);
325                 __d_drop(dentry);
326 #ifdef DCACHE_LUSTRE_INVALID
327                 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
328 #endif
329                 unlock_dentry(dentry);
330                 __d_rehash(dentry, 0); /* avoid taking dcache_lock inside */
331                 spin_unlock(&dcache_lock);
332                 iput(inode);
333                 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
334                        "refc %d\n", de->d_name.len, de->d_name.name, de,
335                        de->d_parent, de->d_inode, atomic_read(&de->d_count));
336                 return dentry;
337         }
338
339         if (last_discon) {
340                 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
341                         "refc %d\n", last_discon, last_discon->d_inode,
342                         atomic_read(&last_discon->d_count));
343                 dget_locked(last_discon);
344                 spin_unlock(&dcache_lock);
345                 d_rehash(de);
346                 d_move(last_discon, de);
347                 iput(inode);
348                 return last_discon;
349         }
350
351         ll_d_add(de, inode);
352
353         spin_unlock(&dcache_lock);
354
355         return de;
356 }
357
358 static int lookup_it_finish(struct ptlrpc_request *request, int offset,
359                             struct lookup_intent *it, void *data)
360 {
361         struct it_cb_data *icbd = data;
362         struct dentry **de = icbd->icbd_childp;
363         struct inode *parent = icbd->icbd_parent;
364         struct ll_sb_info *sbi = ll_i2sbi(parent);
365         struct inode *inode = NULL;
366         int rc;
367
368         /* NB 1 request reference will be taken away by ll_intent_lock()
369          * when I return */
370         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
371                 ENTRY;
372
373                 rc = ll_prep_inode(&inode, request, offset,
374                                    (*de)->d_sb);
375                 if (rc)
376                         RETURN(rc);
377
378                 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
379                        inode, inode->i_ino, inode->i_generation);
380                 md_set_lock_data(sbi->ll_md_exp,
381                                  &it->d.lustre.it_lock_handle, inode);
382
383                 /* We used to query real size from OSTs here, but actually
384                    this is not needed. For stat() calls size would be updated
385                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
386                    2.4 and
387                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
388                    Everybody else who needs correct file size would call
389                    ll_glimpse_size or some equivalent themselves anyway.
390                    Also see bug 7198. */
391
392                 *de = ll_find_alias(inode, *de);
393         } else {
394                 ENTRY;
395                 /* Check that parent has UPDATE lock. If there is none, we
396                    cannot afford to hash this dentry (done by ll_d_add) as it
397                    might get picked up later when UPDATE lock will appear */
398                 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
399                         spin_lock(&dcache_lock);
400                         ll_d_add(*de, inode);
401                         spin_unlock(&dcache_lock);
402                 } else {
403                         (*de)->d_inode = NULL;
404                         /* We do not want to hash the dentry if don`t have a
405                          * lock, but if this dentry is later used in d_move,
406                          * we'd hit uninitialised list head d_hash, so we just
407                          * do this to init d_hash field but leave dentry
408                          * unhashed. (bug 10796). */
409                         d_rehash(*de);
410                         d_drop(*de);
411                 }
412         }
413
414         ll_set_dd(*de);
415         (*de)->d_op = &ll_d_ops;
416
417         RETURN(0);
418 }
419
420 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
421                                    struct lookup_intent *it, int lookup_flags)
422 {
423         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
424         struct dentry *save = dentry, *retval;
425         struct ptlrpc_request *req = NULL;
426         struct md_op_data *op_data;
427         struct it_cb_data icbd;
428         __u32 opc;
429         int rc;
430         ENTRY;
431
432         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
433                 RETURN(ERR_PTR(-ENAMETOOLONG));
434
435         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
436                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
437                parent->i_generation, parent, LL_IT2STR(it));
438
439         if (d_mountpoint(dentry))
440                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
441
442         ll_frob_intent(&it, &lookup_it);
443
444         /* As do_lookup is called before follow_mount, root dentry may be left
445          * not valid, revalidate it here. */
446         if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
447             (it->it_op & (IT_OPEN | IT_CREAT))) {
448                 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
449                 if (rc)
450                         RETURN(ERR_PTR(rc));
451         }
452
453         icbd.icbd_childp = &dentry;
454         icbd.icbd_parent = parent;
455
456         if (it->it_op & IT_CREAT ||
457             (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
458                 opc = LUSTRE_OPC_CREATE;
459         else
460                 opc = LUSTRE_OPC_ANY;
461
462         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
463                                      dentry->d_name.len, lookup_flags, opc,
464                                      NULL);
465         if (IS_ERR(op_data))
466                 RETURN((void *)op_data);
467
468         it->it_create_mode &= ~current->fs->umask;
469
470         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
471                             lookup_flags, &req, ll_md_blocking_ast, 0);
472         ll_finish_md_op_data(op_data);
473         if (rc < 0)
474                 GOTO(out, retval = ERR_PTR(rc));
475
476         rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, &icbd);
477         if (rc != 0) {
478                 ll_intent_release(it);
479                 GOTO(out, retval = ERR_PTR(rc));
480         }
481
482         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
483             !S_ISREG(dentry->d_inode->i_mode) &&
484             !S_ISDIR(dentry->d_inode->i_mode)) {
485                 ll_release_openhandle(dentry, it);
486         }
487         ll_lookup_finish_locks(it, dentry);
488
489         if (dentry == save)
490                 GOTO(out, retval = NULL);
491         else
492                 GOTO(out, retval = dentry);
493  out:
494         if (req)
495                 ptlrpc_req_finished(req);
496         return retval;
497 }
498
499 #ifdef HAVE_VFS_INTENT_PATCHES
500 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
501                                    struct nameidata *nd)
502 {
503         struct dentry *de;
504         ENTRY;
505
506         if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
507                 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
508         else
509                 de = ll_lookup_it(parent, dentry, NULL, 0);
510
511         RETURN(de);
512 }
513 #else
514 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
515                                         int lookup_flags)
516 {
517         struct lookup_intent *it;
518
519         OBD_ALLOC(it, sizeof(*it));
520         if (!it)
521                 return ERR_PTR(-ENOMEM);
522
523         if (lookup_flags & LOOKUP_OPEN) {
524                 it->it_op = IT_OPEN;
525                 if (lookup_flags & LOOKUP_CREATE)
526                         it->it_op |= IT_CREAT;
527                 it->it_create_mode = oit->create_mode;
528                 it->it_flags = oit->flags;
529         } else {
530                 it->it_op = IT_GETATTR;
531         }
532
533 #ifndef HAVE_FILE_IN_STRUCT_INTENT
534                 /* Since there is no way to pass our intent to ll_file_open,
535                  * just check the file is there. Actual open will be done
536                  * in ll_file_open */
537                 if (it->it_op & IT_OPEN)
538                         it->it_op = IT_LOOKUP;
539 #endif
540
541         return it;
542 }
543
544 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
545                                    struct nameidata *nd)
546 {
547         struct dentry *de;
548         ENTRY;
549
550         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
551                 struct lookup_intent *it;
552
553 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
554                 /* Did we came here from failed revalidate just to propagate
555                  * its error? */
556                 if (nd->flags & LOOKUP_OPEN)
557                         if (IS_ERR(nd->intent.open.file))
558                                 RETURN((struct dentry *)nd->intent.open.file);
559 #endif
560
561                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
562                         it = ll_d2d(dentry)->lld_it;
563                         ll_d2d(dentry)->lld_it = NULL;
564                 } else {
565                         it = ll_convert_intent(&nd->intent.open, nd->flags);
566                         if (IS_ERR(it))
567                                 RETURN((struct dentry *)it);
568                 }
569
570                 de = ll_lookup_it(parent, dentry, it, nd->flags);
571                 if (de)
572                         dentry = de;
573                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
574                         if (dentry->d_inode &&
575                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
576 #ifdef HAVE_FILE_IN_STRUCT_INTENT
577                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
578                                         // We cannot call open here as it would
579                                         // deadlock.
580                                         ptlrpc_req_finished(
581                                                        (struct ptlrpc_request *)
582                                                           it->d.lustre.it_data);
583                                 } else {
584                                         struct file *filp;
585                                         nd->intent.open.file->private_data = it;
586                                         filp =lookup_instantiate_filp(nd,dentry,
587                                                                       NULL);
588 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
589 /* 2.6.1[456] have a bug in open_namei() that forgets to check
590  * nd->intent.open.file for error, so we need to return it as lookup's result
591  * instead */
592                                         if (IS_ERR(filp)) {
593                                                 if (de)
594                                                         dput(de);
595                                                 de = (struct dentry *) filp;
596                                         }
597 #endif
598
599                                 }
600 #else /* HAVE_FILE_IN_STRUCT_INTENT */
601                                 /* Release open handle as we have no way to
602                                  * pass it to ll_file_open */
603                                 ll_release_openhandle(dentry, it);
604 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
605                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
606                                 // XXX This can only reliably work on assumption
607                                 // that there are NO hashed negative dentries.
608                                 ll_d2d(dentry)->lld_it = it;
609                                 it = NULL; /* Will be freed in ll_create_nd */
610                                 /* We absolutely depend on ll_create_nd to be
611                                  * called to not leak this intent and possible
612                                  * data attached to it */
613                         }
614                 }
615
616                 if (it) {
617                         ll_intent_release(it);
618                         OBD_FREE(it, sizeof(*it));
619                 }
620         } else {
621                 de = ll_lookup_it(parent, dentry, NULL, 0);
622         }
623
624         RETURN(de);
625 }
626 #endif
627
628 /* We depend on "mode" being set with the proper file type/umask by now */
629 static struct inode *ll_create_node(struct inode *dir, const char *name,
630                                     int namelen, const void *data, int datalen,
631                                     int mode, __u64 extra,
632                                     struct lookup_intent *it)
633 {
634         struct inode *inode = NULL;
635         struct ptlrpc_request *request = NULL;
636         struct ll_sb_info *sbi = ll_i2sbi(dir);
637         int rc;
638         ENTRY;
639
640         LASSERT(it && it->d.lustre.it_disposition);
641
642         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
643         request = it->d.lustre.it_data;
644         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
645         rc = ll_prep_inode(&inode, request, DLM_REPLY_REC_OFF, dir->i_sb);
646         if (rc)
647                 GOTO(out, inode = ERR_PTR(rc));
648
649         LASSERT(list_empty(&inode->i_dentry));
650
651         /* We asked for a lock on the directory, but were granted a
652          * lock on the inode.  Since we finally have an inode pointer,
653          * stuff it in the lock. */
654         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
655                inode, inode->i_ino, inode->i_generation);
656         md_set_lock_data(sbi->ll_md_exp,
657                          &it->d.lustre.it_lock_handle, inode);
658         EXIT;
659  out:
660         ptlrpc_req_finished(request);
661         return inode;
662 }
663
664 /*
665  * By the time this is called, we already have created the directory cache
666  * entry for the new file, but it is so far negative - it has no inode.
667  *
668  * We defer creating the OBD object(s) until open, to keep the intent and
669  * non-intent code paths similar, and also because we do not have the MDS
670  * inode number before calling ll_create_node() (which is needed for LOV),
671  * so we would need to do yet another RPC to the MDS to store the LOV EA
672  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
673  * lmm_size in datalen (the MDS still has code which will handle that).
674  *
675  * If the create succeeds, we fill in the inode information
676  * with d_instantiate().
677  */
678 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
679                         struct lookup_intent *it)
680 {
681         struct inode *inode;
682         int rc = 0;
683         ENTRY;
684
685         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
686                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
687                dir->i_generation, dir, LL_IT2STR(it));
688
689         rc = it_open_error(DISP_OPEN_CREATE, it);
690         if (rc)
691                 RETURN(rc);
692
693         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
694                                NULL, 0, mode, 0, it);
695         if (IS_ERR(inode)) {
696                 RETURN(PTR_ERR(inode));
697         }
698
699         d_instantiate(dentry, inode);
700         /* Negative dentry may be unhashed if parent does not have UPDATE lock,
701          * but some callers, e.g. do_coredump, expect dentry to be hashed after
702          * successful create. Hash it here. */
703         spin_lock(&dcache_lock);
704         if (d_unhashed(dentry))
705                 __d_rehash(dentry, 0);
706         spin_unlock(&dcache_lock);
707         RETURN(0);
708 }
709
710 static void ll_update_times(struct ptlrpc_request *request, int offset,
711                             struct inode *inode)
712 {
713         struct mdt_body *body = lustre_msg_buf(request->rq_repmsg, offset,
714                                                sizeof(*body));
715         LASSERT(body);
716
717         /* mtime is always updated with ctime, but can be set in past.
718            As write and utime(2) may happen within 1 second, and utime's
719            mtime has a priority over write's one, so take mtime from mds
720            for the same ctimes. */
721         if (body->valid & OBD_MD_FLCTIME &&
722             body->ctime >= LTIME_S(inode->i_ctime)) {
723                 LTIME_S(inode->i_ctime) = body->ctime;
724
725                 if (body->valid & OBD_MD_FLMTIME) {
726                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
727                                "to "LPU64"\n", inode->i_ino,
728                                LTIME_S(inode->i_mtime), body->mtime);
729                         LTIME_S(inode->i_mtime) = body->mtime;
730                 }
731         }
732 }
733
734 static int ll_new_node(struct inode *dir, struct qstr *name,
735                        const char *tgt, int mode, int rdev,
736                        struct dentry *dchild, __u32 opc)
737 {
738         struct ptlrpc_request *request = NULL;
739         struct md_op_data *op_data;
740         struct inode *inode = NULL;
741         struct ll_sb_info *sbi = ll_i2sbi(dir);
742         int tgt_len = 0;
743         int err;
744
745         ENTRY;
746         if (unlikely(tgt != NULL))
747                 tgt_len = strlen(tgt) + 1;
748
749         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
750                                      name->len, 0, opc, NULL);
751         if (IS_ERR(op_data))
752                 GOTO(err_exit, err = PTR_ERR(op_data));
753
754         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
755                         current->fsuid, current->fsgid,
756                         current->cap_effective, rdev, &request);
757         ll_finish_md_op_data(op_data);
758         if (err)
759                 GOTO(err_exit, err);
760
761         ll_update_times(request, REPLY_REC_OFF, dir);
762
763         if (dchild) {
764                 err = ll_prep_inode(&inode, request, REPLY_REC_OFF,
765                                     dchild->d_sb);
766                 if (err)
767                      GOTO(err_exit, err);
768
769                 d_drop(dchild);
770                 d_instantiate(dchild, inode);
771                 EXIT;
772         }
773 err_exit:
774         ptlrpc_req_finished(request);
775
776         return err;
777 }
778
779 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
780                             unsigned rdev, struct dentry *dchild)
781 {
782         int err;
783         ENTRY;
784
785         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
786                name->len, name->name, dir->i_ino, dir->i_generation, dir,
787                mode, rdev);
788
789         mode &= ~current->fs->umask;
790
791         switch (mode & S_IFMT) {
792         case 0:
793                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
794         case S_IFREG:
795         case S_IFCHR:
796         case S_IFBLK:
797         case S_IFIFO:
798         case S_IFSOCK:
799                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
800                                   LUSTRE_OPC_MKNOD);
801                 break;
802         case S_IFDIR:
803                 err = -EPERM;
804                 break;
805         default:
806                 err = -EINVAL;
807         }
808         RETURN(err);
809 }
810
811 #ifndef HAVE_VFS_INTENT_PATCHES
812 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
813                         int mode, struct nameidata *nd)
814 {
815         struct lookup_intent *it = ll_d2d(dentry)->lld_it;
816         int rc;
817
818         if (!it)
819                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
820
821         ll_d2d(dentry)->lld_it = NULL;
822
823         /* Was there an error? Propagate it! */
824         if (it->d.lustre.it_status) {
825                 rc = it->d.lustre.it_status;
826                 goto out;
827         }
828
829         rc = ll_create_it(dir, dentry, mode, it);
830 #ifdef HAVE_FILE_IN_STRUCT_INTENT
831         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
832                 nd->intent.open.file->private_data = it;
833                 lookup_instantiate_filp(nd, dentry, NULL);
834         }
835 #else
836         ll_release_openhandle(dentry,it);
837 #endif
838
839 out:
840         ll_intent_release(it);
841         OBD_FREE(it, sizeof(*it));
842
843         return rc;
844 }
845 #else
846 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
847                         int mode, struct nameidata *nd)
848 {
849         if (!nd || !nd->intent.d.lustre.it_disposition)
850                 /* No saved request? Just mknod the file */
851                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
852
853         return ll_create_it(dir, dentry, mode, &nd->intent);
854 }
855 #endif
856
857 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
858                               const char *tgt, struct dentry *dchild)
859 {
860         int err;
861         ENTRY;
862
863         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
864                name->len, name->name, dir->i_ino, dir->i_generation,
865                dir, 3000, tgt);
866
867         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
868                           0, dchild, LUSTRE_OPC_SYMLINK);
869         RETURN(err);
870 }
871
872 static int ll_link_generic(struct inode *src,  struct inode *dir,
873                            struct qstr *name, struct dentry *dchild)
874 {
875         struct ll_sb_info *sbi = ll_i2sbi(dir);
876         struct ptlrpc_request *request = NULL;
877         struct md_op_data *op_data;
878         int err;
879
880         ENTRY;
881         CDEBUG(D_VFSTRACE,
882                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
883                src->i_ino, src->i_generation, src, dir->i_ino,
884                dir->i_generation, dir, name->len, name->name);
885
886         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
887                                      0, LUSTRE_OPC_ANY, NULL);
888         if (IS_ERR(op_data))
889                 RETURN(PTR_ERR(op_data));
890
891         err = md_link(sbi->ll_md_exp, op_data, &request);
892         ll_finish_md_op_data(op_data);
893         if (err)
894                 GOTO(out, err);
895         if (dchild)
896                 d_drop(dchild);
897
898         ll_update_times(request, REPLY_REC_OFF, dir);
899         EXIT;
900 out:
901         ptlrpc_req_finished(request);
902         RETURN(err);
903 }
904
905 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
906                             int mode, struct dentry *dchild)
907
908 {
909         int err;
910         ENTRY;
911
912         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
913                name->len, name->name, dir->i_ino, dir->i_generation, dir);
914
915         mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
916         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
917
918         RETURN(err);
919 }
920
921 /* Try to find the child dentry by its name.
922    If found, put the result fid into @fid. */
923 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
924                              struct lu_fid *fid)
925 {
926         struct dentry *parent, *child;
927         
928         parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
929         child = d_lookup(parent, name);
930         if (child) {
931                 if (child->d_inode)
932                         *fid = *ll_inode2fid(child->d_inode);
933                 dput(child);
934         }
935 }
936
937 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
938                             struct dentry *dchild, struct qstr *name)
939 {
940         struct ptlrpc_request *request = NULL;
941         struct md_op_data *op_data;
942         int rc;
943         ENTRY;
944         
945         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
946                name->len, name->name, dir->i_ino, dir->i_generation, dir);
947
948         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
949                 RETURN(-EBUSY);
950
951         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
952                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
953         if (IS_ERR(op_data))
954                 RETURN(PTR_ERR(op_data));
955
956         ll_get_child_fid(dir, name, &op_data->op_fid3);
957         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
958         ll_finish_md_op_data(op_data);
959         if (rc == 0)
960                 ll_update_times(request, REPLY_REC_OFF, dir);
961         ptlrpc_req_finished(request);
962         RETURN(rc);
963 }
964
965 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
966 {
967         struct mdt_body *body;
968         struct lov_mds_md *eadata;
969         struct lov_stripe_md *lsm = NULL;
970         struct obd_trans_info oti = { 0 };
971         struct obdo *oa;
972         int rc;
973         ENTRY;
974
975         /* req is swabbed so this is safe */
976         body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
977
978         if (!(body->valid & OBD_MD_FLEASIZE))
979                 RETURN(0);
980
981         if (body->eadatasize == 0) {
982                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
983                 GOTO(out, rc = -EPROTO);
984         }
985
986         /* The MDS sent back the EA because we unlinked the last reference
987          * to this file. Use this EA to unlink the objects on the OST.
988          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
989          * check it is complete and sensible. */
990         eadata = lustre_swab_repbuf(request, REPLY_REC_OFF + 1,
991                                     body->eadatasize, NULL);
992         LASSERT(eadata != NULL);
993         if (eadata == NULL) {
994                 CERROR("Can't unpack MDS EA data\n");
995                 GOTO(out, rc = -EPROTO);
996         }
997
998         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
999         if (rc < 0) {
1000                 CERROR("obd_unpackmd: %d\n", rc);
1001                 GOTO(out, rc);
1002         }
1003         LASSERT(rc >= sizeof(*lsm));
1004
1005         rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1006         if (rc)
1007                 GOTO(out_free_memmd, rc);
1008
1009         OBDO_ALLOC(oa);
1010         if (oa == NULL)
1011                 GOTO(out_free_memmd, rc = -ENOMEM);
1012
1013         oa->o_id = lsm->lsm_object_id;
1014         oa->o_gr = lsm->lsm_object_gr;
1015         oa->o_mode = body->mode & S_IFMT;
1016         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1017
1018         if (body->valid & OBD_MD_FLCOOKIE) {
1019                 oa->o_valid |= OBD_MD_FLCOOKIE;
1020                 oti.oti_logcookies =
1021                         lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF + 2,
1022                                        sizeof(struct llog_cookie) *
1023                                        lsm->lsm_stripe_count);
1024                 if (oti.oti_logcookies == NULL) {
1025                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1026                         body->valid &= ~OBD_MD_FLCOOKIE;
1027                 }
1028         }
1029
1030         rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir));
1031         OBDO_FREE(oa);
1032         if (rc)
1033                 CERROR("obd destroy objid "LPX64" error %d\n",
1034                        lsm->lsm_object_id, rc);
1035  out_free_memmd:
1036         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1037  out:
1038         return rc;
1039 }
1040
1041 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1042                              struct dentry *dchild, struct qstr *name)
1043 {
1044         struct ptlrpc_request *request = NULL;
1045         struct md_op_data *op_data;
1046         int rc;
1047         ENTRY;
1048         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1049                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1050
1051         /*
1052          * XXX: unlink bind mountpoint maybe call to here,
1053          * just check it as vfs_unlink does.
1054          */
1055         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1056                 RETURN(-EBUSY);
1057
1058         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1059                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1060         if (IS_ERR(op_data))
1061                 RETURN(PTR_ERR(op_data));
1062
1063         ll_get_child_fid(dir, name, &op_data->op_fid3);
1064         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1065         ll_finish_md_op_data(op_data);
1066         if (rc)
1067                 GOTO(out, rc);
1068
1069         ll_update_times(request, REPLY_REC_OFF, dir);
1070
1071         rc = ll_objects_destroy(request, dir);
1072  out:
1073         ptlrpc_req_finished(request);
1074         RETURN(rc);
1075 }
1076
1077 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1078                              struct dentry *src_dchild, struct qstr *src_name,
1079                              struct inode *tgt, struct dentry *tgt_dparent,
1080                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1081 {
1082         struct ptlrpc_request *request = NULL;
1083         struct ll_sb_info *sbi = ll_i2sbi(src);
1084         struct md_op_data *op_data;
1085         int err;
1086         ENTRY;
1087         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1088                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1089                src->i_ino, src->i_generation, src, tgt_name->len,
1090                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1091
1092         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1093             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1094                 RETURN(-EBUSY);
1095
1096         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1097                                      LUSTRE_OPC_ANY, NULL);
1098         if (IS_ERR(op_data))
1099                 RETURN(PTR_ERR(op_data));
1100
1101         ll_get_child_fid(src, src_name, &op_data->op_fid3);
1102         ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1103         err = md_rename(sbi->ll_md_exp, op_data,
1104                         src_name->name, src_name->len,
1105                         tgt_name->name, tgt_name->len, &request);
1106         ll_finish_md_op_data(op_data);
1107         if (!err) {
1108                 ll_update_times(request, REPLY_REC_OFF, src);
1109                 ll_update_times(request, REPLY_REC_OFF, tgt);
1110                 err = ll_objects_destroy(request, src);
1111         }
1112
1113         ptlrpc_req_finished(request);
1114
1115         RETURN(err);
1116 }
1117
1118 #ifdef HAVE_VFS_INTENT_PATCHES
1119 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1120 {
1121         return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1122 }
1123 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1124 {
1125         return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1126                                  NULL, &srcnd->last,
1127                                  tgtnd->dentry->d_inode, tgtnd->dentry,
1128                                  NULL, &tgtnd->last);
1129 }
1130 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1131 {
1132         return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1133                                &tgtnd->last, NULL);
1134 }
1135 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1136 {
1137         return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1138 }
1139 static int ll_rmdir_raw(struct nameidata *nd)
1140 {
1141         return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1142                                 &nd->last);
1143 }
1144 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1145 {
1146         return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1147 }
1148 static int ll_unlink_raw(struct nameidata *nd)
1149 {
1150         return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1151                                  &nd->last);
1152 }
1153 #endif
1154
1155 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1156                     ll_dev_t rdev)
1157 {
1158         return ll_mknod_generic(dir, &dchild->d_name, mode,
1159                                 old_encode_dev(rdev), dchild);
1160 }
1161
1162 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1163 {
1164         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1165 }
1166 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1167 {
1168         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1169 }
1170 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1171 {
1172         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1173 }
1174 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1175                       const char *oldname)
1176 {
1177         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1178 }
1179 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1180                    struct dentry *new_dentry)
1181 {
1182         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1183                                new_dentry);
1184 }
1185 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1186                      struct inode *new_dir, struct dentry *new_dentry)
1187 {
1188         return ll_rename_generic(old_dir, NULL,
1189                                  old_dentry, &old_dentry->d_name,
1190                                  new_dir, NULL, new_dentry,
1191                                  &new_dentry->d_name);
1192 }
1193
1194 struct inode_operations ll_dir_inode_operations = {
1195 #ifdef HAVE_VFS_INTENT_PATCHES
1196         .link_raw           = ll_link_raw,
1197         .unlink_raw         = ll_unlink_raw,
1198         .symlink_raw        = ll_symlink_raw,
1199         .mkdir_raw          = ll_mkdir_raw,
1200         .rmdir_raw          = ll_rmdir_raw,
1201         .mknod_raw          = ll_mknod_raw,
1202         .rename_raw         = ll_rename_raw,
1203         .setattr            = ll_setattr,
1204         .setattr_raw        = ll_setattr_raw,
1205 #endif
1206         .mknod              = ll_mknod,
1207         .lookup             = ll_lookup_nd,
1208         .create             = ll_create_nd,
1209         /* We need all these non-raw things for NFSD, to not patch it. */
1210         .unlink             = ll_unlink,
1211         .mkdir              = ll_mkdir,
1212         .rmdir              = ll_rmdir,
1213         .symlink            = ll_symlink,
1214         .link               = ll_link,
1215         .rename             = ll_rename,
1216         .setattr            = ll_setattr,
1217         .getattr            = ll_getattr,
1218         .permission         = ll_inode_permission,
1219         .setxattr           = ll_setxattr,
1220         .getxattr           = ll_getxattr,
1221         .listxattr          = ll_listxattr,
1222         .removexattr        = ll_removexattr,
1223 };
1224
1225 struct inode_operations ll_special_inode_operations = {
1226 #ifdef HAVE_VFS_INTENT_PATCHES
1227         .setattr_raw    = ll_setattr_raw,
1228 #endif
1229         .setattr        = ll_setattr,
1230         .getattr        = ll_getattr,
1231         .permission     = ll_inode_permission,
1232         .setxattr       = ll_setxattr,
1233         .getxattr       = ll_getxattr,
1234         .listxattr      = ll_listxattr,
1235         .removexattr    = ll_removexattr,
1236 };