Whamcloud - gitweb
Branch HEAD
[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 LUSTRE_KERNEL_VERSION
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                 __d_rehash(de, 0);
344                 dget_locked(last_discon);
345                 __d_move(last_discon, de);
346                 spin_unlock(&dcache_lock);
347                 d_rehash(de);
348                 iput(inode);
349                 return last_discon;
350         }
351        
352         ll_d_add(de, inode);
353
354         spin_unlock(&dcache_lock);
355
356         return de;
357 }
358
359 static int lookup_it_finish(struct ptlrpc_request *request, int offset,
360                             struct lookup_intent *it, void *data)
361 {
362         struct it_cb_data *icbd = data;
363         struct dentry **de = icbd->icbd_childp;
364         struct inode *parent = icbd->icbd_parent;
365         struct ll_sb_info *sbi = ll_i2sbi(parent);
366         struct inode *inode = NULL;
367         int rc;
368
369         /* NB 1 request reference will be taken away by ll_intent_lock()
370          * when I return */
371         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
372                 ENTRY;
373
374                 rc = ll_prep_inode(&inode, request, offset,
375                                    (*de)->d_sb);
376                 if (rc)
377                         RETURN(rc);
378
379                 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
380                        inode, inode->i_ino, inode->i_generation);
381                 md_set_lock_data(sbi->ll_md_exp,
382                                  &it->d.lustre.it_lock_handle, inode);
383
384                 /* We used to query real size from OSTs here, but actually
385                    this is not needed. For stat() calls size would be updated
386                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
387                    2.4 and
388                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
389                    Everybody else who needs correct file size would call
390                    ll_glimpse_size or some equivalent themselves anyway.
391                    Also see bug 7198. */
392
393                 *de = ll_find_alias(inode, *de);
394         } else {
395                 ENTRY;
396                 /* Check that parent has UPDATE lock. If there is none, we
397                    cannot afford to hash this dentry (done by ll_d_add) as it
398                    might get picked up later when UPDATE lock will appear */
399                 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
400                         spin_lock(&dcache_lock);
401                         ll_d_add(*de, inode);
402                         spin_unlock(&dcache_lock);
403                 } else {
404                         (*de)->d_inode = NULL;
405                         /* We do not want to hash the dentry if don`t have a
406                          * lock, but if this dentry is later used in d_move,
407                          * we'd hit uninitialised list head d_hash, so we just
408                          * do this to init d_hash field but leave dentry
409                          * unhashed. (bug 10796). */
410                         d_rehash(*de);
411                         d_drop(*de);
412                 }
413         }
414
415         ll_set_dd(*de);
416         (*de)->d_op = &ll_d_ops;
417
418         RETURN(0);
419 }
420
421 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
422                                    struct lookup_intent *it, int lookup_flags)
423 {
424         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
425         struct dentry *save = dentry, *retval;
426         struct ptlrpc_request *req = NULL;
427         struct md_op_data *op_data;
428         struct it_cb_data icbd;
429         __u32 opc;
430         int rc;
431         ENTRY;
432
433         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
434                 RETURN(ERR_PTR(-ENAMETOOLONG));
435
436         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
437                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
438                parent->i_generation, parent, LL_IT2STR(it));
439
440         if (d_mountpoint(dentry))
441                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
442
443         ll_frob_intent(&it, &lookup_it);
444
445         /* As do_lookup is called before follow_mount, root dentry may be left
446          * not valid, revalidate it here. */
447         if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
448             (it->it_op & (IT_OPEN | IT_CREAT))) {
449                 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
450                 if (rc)
451                         RETURN(ERR_PTR(rc));
452         }
453
454         icbd.icbd_childp = &dentry;
455         icbd.icbd_parent = parent;
456
457         if (it->it_op & IT_CREAT ||
458             (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
459                 opc = LUSTRE_OPC_CREATE;
460         else
461                 opc = LUSTRE_OPC_ANY;
462
463         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
464                                      dentry->d_name.len, lookup_flags, opc,
465                                      NULL);
466         if (IS_ERR(op_data))
467                 RETURN((void *)op_data);
468
469         it->it_create_mode &= ~current->fs->umask;
470
471         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
472                             lookup_flags, &req, ll_md_blocking_ast, 0);
473         ll_finish_md_op_data(op_data);
474         if (rc < 0)
475                 GOTO(out, retval = ERR_PTR(rc));
476
477         rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, &icbd);
478         if (rc != 0) {
479                 ll_intent_release(it);
480                 GOTO(out, retval = ERR_PTR(rc));
481         }
482
483         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
484             !S_ISREG(dentry->d_inode->i_mode) &&
485             !S_ISDIR(dentry->d_inode->i_mode)) {
486                 ll_release_openhandle(dentry, it);
487         }
488         ll_lookup_finish_locks(it, dentry);
489
490         if (dentry == save)
491                 GOTO(out, retval = NULL);
492         else
493                 GOTO(out, retval = dentry);
494  out:
495         if (req)
496                 ptlrpc_req_finished(req);
497         return retval;
498 }
499
500 #ifdef LUSTRE_KERNEL_VERSION
501 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
502                                    struct nameidata *nd)
503 {
504         struct dentry *de;
505         ENTRY;
506
507         if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
508                 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
509         else
510                 de = ll_lookup_it(parent, dentry, NULL, 0);
511
512         RETURN(de);
513 }
514 #else
515 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
516                                         int lookup_flags)
517 {
518         struct lookup_intent *it;
519
520         OBD_ALLOC(it, sizeof(*it));
521         if (!it)
522                 return ERR_PTR(-ENOMEM);
523
524         if (lookup_flags & LOOKUP_OPEN) {
525                 it->it_op = IT_OPEN;
526                 if (lookup_flags & LOOKUP_CREATE)
527                         it->it_op |= IT_CREAT;
528                 it->it_create_mode = oit->create_mode;
529                 it->it_flags = oit->flags;
530         } else {
531                 it->it_op = IT_GETATTR;
532         }
533
534 #ifndef HAVE_FILE_IN_STRUCT_INTENT
535                 /* Since there is no way to pass our intent to ll_file_open,
536                  * just check the file is there. Actual open will be done
537                  * in ll_file_open */
538                 if (it->it_op & IT_OPEN)
539                         it->it_op = IT_LOOKUP;
540 #endif
541
542         return it;
543 }
544
545 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
546                                    struct nameidata *nd)
547 {
548         struct dentry *de;
549         ENTRY;
550
551         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
552                 struct lookup_intent *it;
553
554 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
555                 /* Did we came here from failed revalidate just to propagate
556                  * its error? */
557                 if (nd->flags & LOOKUP_OPEN)
558                         if (IS_ERR(nd->intent.open.file))
559                                 RETURN((struct dentry *)nd->intent.open.file);
560 #endif
561
562                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
563                         it = ll_d2d(dentry)->lld_it;
564                         ll_d2d(dentry)->lld_it = NULL;
565                 } else {
566                         it = ll_convert_intent(&nd->intent.open, nd->flags);
567                         if (IS_ERR(it))
568                                 RETURN((struct dentry *)it);
569                 }
570
571                 de = ll_lookup_it(parent, dentry, it, nd->flags);
572                 if (de)
573                         dentry = de;
574                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
575                         if (dentry->d_inode &&
576                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
577 #ifdef HAVE_FILE_IN_STRUCT_INTENT
578                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
579                                         // We cannot call open here as it would
580                                         // deadlock.
581                                         ptlrpc_req_finished(
582                                                        (struct ptlrpc_request *)
583                                                           it->d.lustre.it_data);
584                                 } else {
585                                         struct file *filp;
586                                         nd->intent.open.file->private_data = it;
587                                         filp =lookup_instantiate_filp(nd,dentry,
588                                                                       NULL);
589 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
590 /* 2.6.1[456] have a bug in open_namei() that forgets to check
591  * nd->intent.open.file for error, so we need to return it as lookup's result
592  * instead */
593                                         if (IS_ERR(filp)) {
594                                                 if (de)
595                                                         dput(de);
596                                                 de = (struct dentry *) filp;
597                                         }
598 #endif
599
600                                 }
601 #else /* HAVE_FILE_IN_STRUCT_INTENT */
602                                 /* Release open handle as we have no way to
603                                  * pass it to ll_file_open */
604                                 ll_release_openhandle(dentry, it);
605 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
606                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
607                                 // XXX This can only reliably work on assumption
608                                 // that there are NO hashed negative dentries.
609                                 ll_d2d(dentry)->lld_it = it;
610                                 it = NULL; /* Will be freed in ll_create_nd */
611                                 /* We absolutely depend on ll_create_nd to be
612                                  * called to not leak this intent and possible
613                                  * data attached to it */
614                         }
615                 }
616
617                 if (it) {
618                         ll_intent_release(it);
619                         OBD_FREE(it, sizeof(*it));
620                 }
621         } else {
622                 de = ll_lookup_it(parent, dentry, NULL, 0);
623         }
624
625         RETURN(de);
626 }
627 #endif
628
629 /* We depend on "mode" being set with the proper file type/umask by now */
630 static struct inode *ll_create_node(struct inode *dir, const char *name,
631                                     int namelen, const void *data, int datalen,
632                                     int mode, __u64 extra,
633                                     struct lookup_intent *it)
634 {
635         struct inode *inode = NULL;
636         struct ptlrpc_request *request = NULL;
637         struct ll_sb_info *sbi = ll_i2sbi(dir);
638         int rc;
639         ENTRY;
640
641         LASSERT(it && it->d.lustre.it_disposition);
642
643         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
644         request = it->d.lustre.it_data;
645         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
646         rc = ll_prep_inode(&inode, request, DLM_REPLY_REC_OFF, dir->i_sb);
647         if (rc)
648                 GOTO(out, inode = ERR_PTR(rc));
649
650         LASSERT(list_empty(&inode->i_dentry));
651
652         /* We asked for a lock on the directory, but were granted a
653          * lock on the inode.  Since we finally have an inode pointer,
654          * stuff it in the lock. */
655         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
656                inode, inode->i_ino, inode->i_generation);
657         md_set_lock_data(sbi->ll_md_exp,
658                          &it->d.lustre.it_lock_handle, inode);
659         EXIT;
660  out:
661         ptlrpc_req_finished(request);
662         return inode;
663 }
664
665 /*
666  * By the time this is called, we already have created the directory cache
667  * entry for the new file, but it is so far negative - it has no inode.
668  *
669  * We defer creating the OBD object(s) until open, to keep the intent and
670  * non-intent code paths similar, and also because we do not have the MDS
671  * inode number before calling ll_create_node() (which is needed for LOV),
672  * so we would need to do yet another RPC to the MDS to store the LOV EA
673  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
674  * lmm_size in datalen (the MDS still has code which will handle that).
675  *
676  * If the create succeeds, we fill in the inode information
677  * with d_instantiate().
678  */
679 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
680                         struct lookup_intent *it)
681 {
682         struct inode *inode;
683         int rc = 0;
684         ENTRY;
685
686         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
687                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
688                dir->i_generation, dir, LL_IT2STR(it));
689
690         rc = it_open_error(DISP_OPEN_CREATE, it);
691         if (rc)
692                 RETURN(rc);
693
694         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
695                                NULL, 0, mode, 0, it);
696         if (IS_ERR(inode)) {
697                 RETURN(PTR_ERR(inode));
698         }
699
700         d_instantiate(dentry, inode);
701         /* Negative dentry may be unhashed if parent does not have UPDATE lock,
702          * but some callers, e.g. do_coredump, expect dentry to be hashed after
703          * successful create. Hash it here. */
704         spin_lock(&dcache_lock);
705         if (d_unhashed(dentry))
706                 __d_rehash(dentry, 0);
707         spin_unlock(&dcache_lock);
708         RETURN(0);
709 }
710
711 static void ll_update_times(struct ptlrpc_request *request, int offset,
712                             struct inode *inode)
713 {
714         struct mdt_body *body = lustre_msg_buf(request->rq_repmsg, offset,
715                                                sizeof(*body));
716         LASSERT(body);
717
718         /* mtime is always updated with ctime, but can be set in past.
719            As write and utime(2) may happen within 1 second, and utime's
720            mtime has a priority over write's one, so take mtime from mds
721            for the same ctimes. */
722         if (body->valid & OBD_MD_FLCTIME &&
723             body->ctime >= LTIME_S(inode->i_ctime)) {
724                 LTIME_S(inode->i_ctime) = body->ctime;
725
726                 if (body->valid & OBD_MD_FLMTIME) {
727                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
728                                "to "LPU64"\n", inode->i_ino,
729                                LTIME_S(inode->i_mtime), body->mtime);
730                         LTIME_S(inode->i_mtime) = body->mtime;
731                 }
732         }
733 }
734
735 static int ll_new_node(struct inode *dir, struct qstr *name,
736                        const char *tgt, int mode, int rdev,
737                        struct dentry *dchild, __u32 opc)
738 {
739         struct ptlrpc_request *request = NULL;
740         struct md_op_data *op_data;
741         struct inode *inode = NULL;
742         struct ll_sb_info *sbi = ll_i2sbi(dir);
743         int tgt_len = 0;
744         int err;
745
746         ENTRY;
747         if (unlikely(tgt != NULL))
748                 tgt_len = strlen(tgt) + 1;
749
750         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
751                                      name->len, 0, opc, NULL);
752         if (IS_ERR(op_data))
753                 GOTO(err_exit, err = PTR_ERR(op_data));
754
755         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
756                         current->fsuid, current->fsgid,
757                         current->cap_effective, rdev, &request);
758         ll_finish_md_op_data(op_data);
759         if (err)
760                 GOTO(err_exit, err);
761
762         ll_update_times(request, REPLY_REC_OFF, dir);
763
764         if (dchild) {
765                 err = ll_prep_inode(&inode, request, REPLY_REC_OFF,
766                                     dchild->d_sb);
767                 if (err)
768                      GOTO(err_exit, err);
769
770                 d_drop(dchild);
771                 d_instantiate(dchild, inode);
772                 EXIT;
773         }
774 err_exit:
775         ptlrpc_req_finished(request);
776
777         return err;
778 }
779
780 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
781                             unsigned rdev, struct dentry *dchild)
782 {
783         int err;
784         ENTRY;
785
786         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
787                name->len, name->name, dir->i_ino, dir->i_generation, dir,
788                mode, rdev);
789
790         mode &= ~current->fs->umask;
791
792         switch (mode & S_IFMT) {
793         case 0:
794                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
795         case S_IFREG:
796         case S_IFCHR:
797         case S_IFBLK:
798         case S_IFIFO:
799         case S_IFSOCK:
800                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
801                                   LUSTRE_OPC_MKNOD);
802                 break;
803         case S_IFDIR:
804                 err = -EPERM;
805                 break;
806         default:
807                 err = -EINVAL;
808         }
809         RETURN(err);
810 }
811
812 #ifndef LUSTRE_KERNEL_VERSION
813 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
814                         int mode, struct nameidata *nd)
815 {
816         struct lookup_intent *it = ll_d2d(dentry)->lld_it;
817         int rc;
818
819         if (!it)
820                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
821
822         ll_d2d(dentry)->lld_it = NULL;
823
824         /* Was there an error? Propagate it! */
825         if (it->d.lustre.it_status) {
826                 rc = it->d.lustre.it_status;
827                 goto out;
828         }
829
830         rc = ll_create_it(dir, dentry, mode, it);
831 #ifdef HAVE_FILE_IN_STRUCT_INTENT
832         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
833                 nd->intent.open.file->private_data = it;
834                 lookup_instantiate_filp(nd, dentry, NULL);
835         }
836 #else
837         ll_release_openhandle(dentry,it);
838 #endif
839
840 out:
841         ll_intent_release(it);
842         OBD_FREE(it, sizeof(*it));
843
844         return rc;
845 }
846 #else
847 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
848                         int mode, struct nameidata *nd)
849 {
850         if (!nd || !nd->intent.d.lustre.it_disposition)
851                 /* No saved request? Just mknod the file */
852                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
853
854         return ll_create_it(dir, dentry, mode, &nd->intent);
855 }
856 #endif
857
858 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
859                               const char *tgt, struct dentry *dchild)
860 {
861         int err;
862         ENTRY;
863
864         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
865                name->len, name->name, dir->i_ino, dir->i_generation,
866                dir, 3000, tgt);
867
868         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
869                           0, dchild, LUSTRE_OPC_SYMLINK);
870         RETURN(err);
871 }
872
873 static int ll_link_generic(struct inode *src,  struct inode *dir,
874                            struct qstr *name, struct dentry *dchild)
875 {
876         struct ll_sb_info *sbi = ll_i2sbi(dir);
877         struct ptlrpc_request *request = NULL;
878         struct md_op_data *op_data;
879         int err;
880
881         ENTRY;
882         CDEBUG(D_VFSTRACE,
883                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
884                src->i_ino, src->i_generation, src, dir->i_ino,
885                dir->i_generation, dir, name->len, name->name);
886
887         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
888                                      0, LUSTRE_OPC_ANY, NULL);
889         if (IS_ERR(op_data))
890                 RETURN(PTR_ERR(op_data));
891
892         err = md_link(sbi->ll_md_exp, op_data, &request);
893         ll_finish_md_op_data(op_data);
894         if (err)
895                 GOTO(out, err);
896         if (dchild)
897                 d_drop(dchild);
898
899         ll_update_times(request, REPLY_REC_OFF, dir);
900         EXIT;
901 out:
902         ptlrpc_req_finished(request);
903         RETURN(err);
904 }
905
906 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
907                             int mode, struct dentry *dchild)
908
909 {
910         int err;
911         ENTRY;
912
913         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
914                name->len, name->name, dir->i_ino, dir->i_generation, dir);
915
916         mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
917         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
918
919         RETURN(err);
920 }
921
922 /* Try to find the child dentry by its name.
923    If found, put the result fid into @fid. */
924 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
925                              struct lu_fid *fid)
926 {
927         struct dentry *parent, *child;
928         
929         parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
930         child = d_lookup(parent, name);
931         if (child) {
932                 if (child->d_inode)
933                         *fid = *ll_inode2fid(child->d_inode);
934                 dput(child);
935         }
936 }
937
938 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
939                             struct dentry *dchild, struct qstr *name)
940 {
941         struct ptlrpc_request *request = NULL;
942         struct md_op_data *op_data;
943         int rc;
944         ENTRY;
945         
946         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
947                name->len, name->name, dir->i_ino, dir->i_generation, dir);
948
949         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
950                 RETURN(-EBUSY);
951
952         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
953                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
954         if (IS_ERR(op_data))
955                 RETURN(PTR_ERR(op_data));
956
957         ll_get_child_fid(dir, name, &op_data->op_fid3);
958         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
959         ll_finish_md_op_data(op_data);
960         if (rc == 0)
961                 ll_update_times(request, REPLY_REC_OFF, dir);
962         ptlrpc_req_finished(request);
963         RETURN(rc);
964 }
965
966 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
967 {
968         struct mdt_body *body;
969         struct lov_mds_md *eadata;
970         struct lov_stripe_md *lsm = NULL;
971         struct obd_trans_info oti = { 0 };
972         struct obdo *oa;
973         int rc;
974         ENTRY;
975
976         /* req is swabbed so this is safe */
977         body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
978
979         if (!(body->valid & OBD_MD_FLEASIZE))
980                 RETURN(0);
981
982         if (body->eadatasize == 0) {
983                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
984                 GOTO(out, rc = -EPROTO);
985         }
986
987         /* The MDS sent back the EA because we unlinked the last reference
988          * to this file. Use this EA to unlink the objects on the OST.
989          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
990          * check it is complete and sensible. */
991         eadata = lustre_swab_repbuf(request, REPLY_REC_OFF + 1,
992                                     body->eadatasize, NULL);
993         LASSERT(eadata != NULL);
994         if (eadata == NULL) {
995                 CERROR("Can't unpack MDS EA data\n");
996                 GOTO(out, rc = -EPROTO);
997         }
998
999         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1000         if (rc < 0) {
1001                 CERROR("obd_unpackmd: %d\n", rc);
1002                 GOTO(out, rc);
1003         }
1004         LASSERT(rc >= sizeof(*lsm));
1005
1006         rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1007         if (rc)
1008                 GOTO(out_free_memmd, rc);
1009
1010         OBDO_ALLOC(oa);
1011         if (oa == NULL)
1012                 GOTO(out_free_memmd, rc = -ENOMEM);
1013
1014         oa->o_id = lsm->lsm_object_id;
1015         oa->o_gr = lsm->lsm_object_gr;
1016         oa->o_mode = body->mode & S_IFMT;
1017         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1018
1019         if (body->valid & OBD_MD_FLCOOKIE) {
1020                 oa->o_valid |= OBD_MD_FLCOOKIE;
1021                 oti.oti_logcookies =
1022                         lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF + 2,
1023                                        sizeof(struct llog_cookie) *
1024                                        lsm->lsm_stripe_count);
1025                 if (oti.oti_logcookies == NULL) {
1026                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1027                         body->valid &= ~OBD_MD_FLCOOKIE;
1028                 }
1029         }
1030
1031         rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir));
1032         OBDO_FREE(oa);
1033         if (rc)
1034                 CERROR("obd destroy objid "LPX64" error %d\n",
1035                        lsm->lsm_object_id, rc);
1036  out_free_memmd:
1037         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1038  out:
1039         return rc;
1040 }
1041
1042 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1043                              struct dentry *dchild, struct qstr *name)
1044 {
1045         struct ptlrpc_request *request = NULL;
1046         struct md_op_data *op_data;
1047         int rc;
1048         ENTRY;
1049         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1050                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1051
1052         /*
1053          * XXX: unlink bind mountpoint maybe call to here,
1054          * just check it as vfs_unlink does.
1055          */
1056         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1057                 RETURN(-EBUSY);
1058
1059         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1060                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1061         if (IS_ERR(op_data))
1062                 RETURN(PTR_ERR(op_data));
1063
1064         ll_get_child_fid(dir, name, &op_data->op_fid3);
1065         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1066         ll_finish_md_op_data(op_data);
1067
1068         if (rc)
1069                 GOTO(out, rc);
1070
1071         ll_update_times(request, REPLY_REC_OFF, dir);
1072
1073         rc = ll_objects_destroy(request, dir);
1074  out:
1075         ptlrpc_req_finished(request);
1076         RETURN(rc);
1077 }
1078
1079 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1080                              struct dentry *src_dchild, struct qstr *src_name,
1081                              struct inode *tgt, struct dentry *tgt_dparent,
1082                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1083 {
1084         struct ptlrpc_request *request = NULL;
1085         struct ll_sb_info *sbi = ll_i2sbi(src);
1086         struct md_op_data *op_data;
1087         int err;
1088         ENTRY;
1089         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1090                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1091                src->i_ino, src->i_generation, src, tgt_name->len,
1092                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1093
1094         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1095             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1096                 RETURN(-EBUSY);
1097
1098         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1099                                      LUSTRE_OPC_ANY, NULL);
1100         if (IS_ERR(op_data))
1101                 RETURN(PTR_ERR(op_data));
1102
1103         ll_get_child_fid(src, src_name, &op_data->op_fid3);
1104         ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1105         err = md_rename(sbi->ll_md_exp, op_data,
1106                         src_name->name, src_name->len,
1107                         tgt_name->name, tgt_name->len, &request);
1108         ll_finish_md_op_data(op_data);
1109         if (!err) {
1110                 ll_update_times(request, REPLY_REC_OFF, src);
1111                 ll_update_times(request, REPLY_REC_OFF, tgt);
1112                 err = ll_objects_destroy(request, src);
1113         }
1114
1115         ptlrpc_req_finished(request);
1116
1117         RETURN(err);
1118 }
1119
1120 #ifdef LUSTRE_KERNEL_VERSION
1121 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1122 {
1123         return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1124 }
1125 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1126 {
1127         return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1128                                  NULL, &srcnd->last,
1129                                  tgtnd->dentry->d_inode, tgtnd->dentry,
1130                                  NULL, &tgtnd->last);
1131 }
1132 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1133 {
1134         return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1135                                &tgtnd->last, NULL);
1136 }
1137 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1138 {
1139         return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1140 }
1141 static int ll_rmdir_raw(struct nameidata *nd)
1142 {
1143         return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1144                                 &nd->last);
1145 }
1146 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1147 {
1148         return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1149 }
1150 static int ll_unlink_raw(struct nameidata *nd)
1151 {
1152         return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1153                                  &nd->last);
1154 }
1155 #endif
1156
1157 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1158                     ll_dev_t rdev)
1159 {
1160         return ll_mknod_generic(dir, &dchild->d_name, mode,
1161                                 old_encode_dev(rdev), dchild);
1162 }
1163
1164 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1165 {
1166         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1167 }
1168 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1169 {
1170         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1171 }
1172 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1173 {
1174         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1175 }
1176 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1177                       const char *oldname)
1178 {
1179         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1180 }
1181 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1182                    struct dentry *new_dentry)
1183 {
1184         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1185                                new_dentry);
1186 }
1187 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1188                      struct inode *new_dir, struct dentry *new_dentry)
1189 {
1190         return ll_rename_generic(old_dir, NULL,
1191                                  old_dentry, &old_dentry->d_name,
1192                                  new_dir, NULL, new_dentry,
1193                                  &new_dentry->d_name);
1194 }
1195
1196 struct inode_operations ll_dir_inode_operations = {
1197 #ifdef LUSTRE_KERNEL_VERSION
1198         .link_raw           = ll_link_raw,
1199         .unlink_raw         = ll_unlink_raw,
1200         .symlink_raw        = ll_symlink_raw,
1201         .mkdir_raw          = ll_mkdir_raw,
1202         .rmdir_raw          = ll_rmdir_raw,
1203         .mknod_raw          = ll_mknod_raw,
1204         .rename_raw         = ll_rename_raw,
1205         .setattr            = ll_setattr,
1206         .setattr_raw        = ll_setattr_raw,
1207 #endif
1208         .mknod              = ll_mknod,
1209         .lookup             = ll_lookup_nd,
1210         .create             = ll_create_nd,
1211         /* We need all these non-raw things for NFSD, to not patch it. */
1212         .unlink             = ll_unlink,
1213         .mkdir              = ll_mkdir,
1214         .rmdir              = ll_rmdir,
1215         .symlink            = ll_symlink,
1216         .link               = ll_link,
1217         .rename             = ll_rename,
1218         .setattr            = ll_setattr,
1219         .getattr            = ll_getattr,
1220         .permission         = ll_inode_permission,
1221         .setxattr           = ll_setxattr,
1222         .getxattr           = ll_getxattr,
1223         .listxattr          = ll_listxattr,
1224         .removexattr        = ll_removexattr,
1225 };
1226
1227 struct inode_operations ll_special_inode_operations = {
1228 #ifdef LUSTRE_KERNEL_VERSION
1229         .setattr_raw    = ll_setattr_raw,
1230 #endif
1231         .setattr        = ll_setattr,
1232         .getattr        = ll_getattr,
1233         .permission     = ll_inode_permission,
1234         .setxattr       = ll_setxattr,
1235         .getxattr       = ll_getxattr,
1236         .listxattr      = ll_listxattr,
1237         .removexattr    = ll_removexattr,
1238 };