Whamcloud - gitweb
e73f203851822357ba9010d2972c01c4562cfcb2
[fs/lustre-release.git] / lustre / llite / namei.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  */
36
37 #include <linux/fs.h>
38 #include <linux/sched.h>
39 #include <linux/mm.h>
40 #include <linux/quotaops.h>
41 #include <linux/highmem.h>
42 #include <linux/pagemap.h>
43 #include <linux/security.h>
44
45 #define DEBUG_SUBSYSTEM S_LLITE
46
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <lustre_lite.h>
50 #include <lustre_dlm.h>
51 #include <lustre_ver.h>
52 #include "llite_internal.h"
53
54 /*
55  * Check if we have something mounted at the named dchild.
56  * In such a case there would always be dentry present.
57  */
58 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
59                            struct qstr *name)
60 {
61         int mounted = 0;
62
63         if (unlikely(dchild)) {
64                 mounted = d_mountpoint(dchild);
65         } else if (dparent) {
66                 dchild = d_lookup(dparent, name);
67                 if (dchild) {
68                         mounted = d_mountpoint(dchild);
69                         dput(dchild);
70                 }
71         }
72         return mounted;
73 }
74
75 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
76 {
77         ENTRY;
78
79         ldlm_lock_decref(lockh, mode);
80
81         RETURN(0);
82 }
83
84
85 /* called from iget5_locked->find_inode() under inode_lock spinlock */
86 static int ll_test_inode(struct inode *inode, void *opaque)
87 {
88         struct ll_inode_info *lli = ll_i2info(inode);
89         struct lustre_md     *md = opaque;
90
91         if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
92                 CERROR("MDS body missing FID\n");
93                 return 0;
94         }
95
96         if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
97                 return 0;
98
99         return 1;
100 }
101
102 static int ll_set_inode(struct inode *inode, void *opaque)
103 {
104         struct ll_inode_info *lli = ll_i2info(inode);
105         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
106
107         if (unlikely(!(body->valid & OBD_MD_FLID))) {
108                 CERROR("MDS body missing FID\n");
109                 return -EINVAL;
110         }
111
112         lli->lli_fid = body->fid1;
113         if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
114                 CERROR("Can not initialize inode "DFID" without object type: "
115                        "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
116                 return -EINVAL;
117         }
118
119         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
120         if (unlikely(inode->i_mode == 0)) {
121                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
122                 return -EINVAL;
123         }
124
125         ll_lli_init(lli);
126
127         return 0;
128 }
129
130
131 /*
132  * Get an inode by inode number (already instantiated by the intent lookup).
133  * Returns inode or NULL
134  */
135 struct inode *ll_iget(struct super_block *sb, ino_t hash,
136                       struct lustre_md *md)
137 {
138         struct inode         *inode;
139         ENTRY;
140
141         LASSERT(hash != 0);
142         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
143
144         if (inode) {
145                 if (inode->i_state & I_NEW) {
146                         int rc = 0;
147
148                         ll_read_inode2(inode, md);
149                         if (S_ISREG(inode->i_mode) &&
150                             ll_i2info(inode)->lli_clob == NULL)
151                                 rc = cl_file_inode_init(inode, md);
152                         if (rc != 0) {
153                                 md->lsm = NULL;
154                                 make_bad_inode(inode);
155                                 unlock_new_inode(inode);
156                                 iput(inode);
157                                 inode = ERR_PTR(rc);
158                         } else
159                                 unlock_new_inode(inode);
160                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
161                         ll_update_inode(inode, md);
162                 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
163                        inode, PFID(&md->body->fid1));
164         }
165         RETURN(inode);
166 }
167
168 static void ll_invalidate_negative_children(struct inode *dir)
169 {
170         struct dentry *dentry, *tmp_subdir;
171
172         ll_lock_dcache(dir);
173         list_for_each_entry(dentry, &dir->i_dentry, d_alias) {
174                 spin_lock(&dentry->d_lock);
175                 if (!list_empty(&dentry->d_subdirs)) {
176                         struct dentry *child;
177
178                         list_for_each_entry_safe(child, tmp_subdir,
179                                                  &dentry->d_subdirs,
180                                                  d_child) {
181                                 if (child->d_inode == NULL)
182                                         d_lustre_invalidate(child);
183                         }
184                 }
185                 spin_unlock(&dentry->d_lock);
186         }
187         ll_unlock_dcache(dir);
188 }
189
190 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
191                        void *data, int flag)
192 {
193         int rc;
194         struct lustre_handle lockh;
195         ENTRY;
196
197         switch (flag) {
198         case LDLM_CB_BLOCKING:
199                 ldlm_lock2handle(lock, &lockh);
200                 rc = ldlm_cli_cancel(&lockh);
201                 if (rc < 0) {
202                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
203                         RETURN(rc);
204                 }
205                 break;
206         case LDLM_CB_CANCELING: {
207                 struct inode *inode = ll_inode_from_lock(lock);
208                 struct ll_inode_info *lli;
209                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
210                 struct lu_fid *fid;
211                 ldlm_mode_t mode = lock->l_req_mode;
212
213                 /* Invalidate all dentries associated with this inode */
214                 if (inode == NULL)
215                         break;
216
217                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
218                 /* For OPEN locks we differentiate between lock modes
219                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
220                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
221                             MDS_INODELOCK_LAYOUT))
222                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
223
224                 if (bits & MDS_INODELOCK_OPEN)
225                         ll_have_md_lock(inode, &bits, mode);
226
227                 fid = ll_inode2fid(inode);
228                 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
229                     lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
230                     lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
231                         LDLM_ERROR(lock, "data mismatch with object "
232                                    DFID" (%p)", PFID(fid), inode);
233                 }
234
235                 if (bits & MDS_INODELOCK_OPEN) {
236                         int flags = 0;
237                         switch (lock->l_req_mode) {
238                         case LCK_CW:
239                                 flags = FMODE_WRITE;
240                                 break;
241                         case LCK_PR:
242                                 flags = FMODE_EXEC;
243                                 break;
244                         case LCK_CR:
245                                 flags = FMODE_READ;
246                                 break;
247                         default:
248                                 CERROR("Unexpected lock mode for OPEN lock "
249                                        "%d, inode %ld\n", lock->l_req_mode,
250                                        inode->i_ino);
251                         }
252                         ll_md_real_close(inode, flags);
253                 }
254
255                 lli = ll_i2info(inode);
256                 if (bits & MDS_INODELOCK_LAYOUT) {
257                         struct cl_object_conf conf = { .coc_inode = inode,
258                                                        .coc_invalidate = true };
259                         rc = ll_layout_conf(inode, &conf);
260                         if (rc)
261                                 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
262                 }
263
264                 if (bits & MDS_INODELOCK_UPDATE)
265                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
266
267                 if (S_ISDIR(inode->i_mode) &&
268                      (bits & MDS_INODELOCK_UPDATE)) {
269                         CDEBUG(D_INODE, "invalidating inode %lu\n",
270                                inode->i_ino);
271                         truncate_inode_pages(inode->i_mapping, 0);
272                         ll_invalidate_negative_children(inode);
273                 }
274
275                 if (inode->i_sb->s_root &&
276                     inode != inode->i_sb->s_root->d_inode &&
277                     (bits & MDS_INODELOCK_LOOKUP))
278                         ll_invalidate_aliases(inode);
279                 iput(inode);
280                 break;
281         }
282         default:
283                 LBUG();
284         }
285
286         RETURN(0);
287 }
288
289 __u32 ll_i2suppgid(struct inode *i)
290 {
291         if (cfs_curproc_is_in_groups(i->i_gid))
292                 return (__u32)i->i_gid;
293         else
294                 return (__u32)(-1);
295 }
296
297 /* Pack the required supplementary groups into the supplied groups array.
298  * If we don't need to use the groups from the target inode(s) then we
299  * instead pack one or more groups from the user's supplementary group
300  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
301 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
302 {
303 #if 0
304         int i;
305 #endif
306
307         LASSERT(i1 != NULL);
308         LASSERT(suppgids != NULL);
309
310         suppgids[0] = ll_i2suppgid(i1);
311
312         if (i2)
313                 suppgids[1] = ll_i2suppgid(i2);
314                 else
315                         suppgids[1] = -1;
316
317 #if 0
318         for (i = 0; i < current_ngroups; i++) {
319                 if (suppgids[0] == -1) {
320                         if (current_groups[i] != suppgids[1])
321                                 suppgids[0] = current_groups[i];
322                         continue;
323                 }
324                 if (suppgids[1] == -1) {
325                         if (current_groups[i] != suppgids[0])
326                                 suppgids[1] = current_groups[i];
327                         continue;
328                 }
329                 break;
330         }
331 #endif
332 }
333
334 /*
335  * try to reuse three types of dentry:
336  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
337  *    by concurrent .revalidate).
338  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
339  *    be cleared by others calling d_lustre_revalidate).
340  * 3. DISCONNECTED alias.
341  */
342 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
343 {
344         struct dentry *alias, *discon_alias, *invalid_alias;
345
346         if (list_empty(&inode->i_dentry))
347                 return NULL;
348
349         discon_alias = invalid_alias = NULL;
350
351         ll_lock_dcache(inode);
352         list_for_each_entry(alias, &inode->i_dentry, d_alias) {
353                 LASSERT(alias != dentry);
354
355                 spin_lock(&alias->d_lock);
356                 if (alias->d_flags & DCACHE_DISCONNECTED)
357                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
358                         discon_alias = alias;
359                 else if (alias->d_parent == dentry->d_parent             &&
360                          alias->d_name.hash == dentry->d_name.hash       &&
361                          alias->d_name.len == dentry->d_name.len         &&
362                          memcmp(alias->d_name.name, dentry->d_name.name,
363                                 dentry->d_name.len) == 0)
364                         invalid_alias = alias;
365                 spin_unlock(&alias->d_lock);
366
367                 if (invalid_alias)
368                         break;
369         }
370         alias = invalid_alias ?: discon_alias ?: NULL;
371         if (alias) {
372                 spin_lock(&alias->d_lock);
373                 dget_dlock(alias);
374                 spin_unlock(&alias->d_lock);
375         }
376         ll_unlock_dcache(inode);
377
378         return alias;
379 }
380
381 /*
382  * Similar to d_splice_alias(), but lustre treats DCACHE_LUSTRE_INVALID alias
383  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
384  */
385 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
386 {
387         struct dentry *new;
388
389         if (inode) {
390                 new = ll_find_alias(inode, de);
391                 if (new) {
392                         ll_dops_init(new, 1, 1);
393                         d_move(new, de);
394                         iput(inode);
395                         CDEBUG(D_DENTRY,
396                                "Reuse dentry %p inode %p refc %d flags %#x\n",
397                               new, new->d_inode, d_refcount(new), new->d_flags);
398                         return new;
399                 }
400         }
401         __d_lustre_invalidate(de);
402         ll_dops_init(de, 1, 1);
403         d_add(de, inode);
404         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
405                de, de->d_inode, d_refcount(de), de->d_flags);
406         return de;
407 }
408
409 int ll_lookup_it_finish(struct ptlrpc_request *request,
410                         struct lookup_intent *it, void *data)
411 {
412         struct it_cb_data *icbd = data;
413         struct dentry **de = icbd->icbd_childp;
414         struct inode *parent = icbd->icbd_parent;
415         struct inode *inode = NULL;
416         __u64 bits = 0;
417         int rc;
418         ENTRY;
419
420         /* NB 1 request reference will be taken away by ll_intent_lock()
421          * when I return */
422         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
423                 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
424                 if (rc)
425                         RETURN(rc);
426
427                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
428
429                 /* We used to query real size from OSTs here, but actually
430                    this is not needed. For stat() calls size would be updated
431                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
432                    2.4 and
433                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
434                    Everybody else who needs correct file size would call
435                    ll_glimpse_size or some equivalent themselves anyway.
436                    Also see bug 7198. */
437         }
438
439         *de = ll_splice_alias(inode, *de);
440
441         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
442                 /* we have lookup look - unhide dentry */
443                 if (bits & MDS_INODELOCK_LOOKUP)
444                         d_lustre_revalidate(*de);
445         } else {
446                 /* Check that parent has UPDATE lock. */
447                 struct lookup_intent parent_it = {
448                                         .it_op = IT_GETATTR,
449                                         .d.lustre.it_lock_handle = 0 };
450
451                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
452                                        &ll_i2info(parent)->lli_fid, NULL)) {
453                         d_lustre_revalidate(*de);
454                         ll_intent_release(&parent_it);
455                 }
456         }
457
458         RETURN(0);
459 }
460
461 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
462                                    struct lookup_intent *it, int lookup_flags)
463 {
464         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
465         struct dentry *save = dentry, *retval;
466         struct ptlrpc_request *req = NULL;
467         struct md_op_data *op_data;
468         struct it_cb_data icbd;
469         __u32 opc;
470         int rc;
471         ENTRY;
472
473         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
474                 RETURN(ERR_PTR(-ENAMETOOLONG));
475
476         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
477                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
478                parent->i_generation, parent, LL_IT2STR(it));
479
480         if (d_mountpoint(dentry))
481                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
482
483         ll_frob_intent(&it, &lookup_it);
484
485         /* As do_lookup is called before follow_mount, root dentry may be left
486          * not valid, revalidate it here. */
487         if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
488             (it->it_op & (IT_OPEN | IT_CREAT))) {
489                 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
490                                             MDS_INODELOCK_LOOKUP);
491                 if (rc)
492                         RETURN(ERR_PTR(rc));
493         }
494
495         if (it->it_op == IT_GETATTR) {
496                 rc = ll_statahead_enter(parent, &dentry, 0);
497                 if (rc == 1) {
498                         if (dentry == save)
499                                 GOTO(out, retval = NULL);
500                         GOTO(out, retval = dentry);
501                 }
502         }
503
504         icbd.icbd_childp = &dentry;
505         icbd.icbd_parent = parent;
506
507         if (it->it_op & IT_CREAT ||
508             (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
509                 opc = LUSTRE_OPC_CREATE;
510         else
511                 opc = LUSTRE_OPC_ANY;
512
513         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
514                                      dentry->d_name.len, lookup_flags, opc,
515                                      NULL);
516         if (IS_ERR(op_data))
517                 RETURN((void *)op_data);
518
519         it->it_create_mode &= ~cfs_curproc_umask();
520
521         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
522                             lookup_flags, &req, ll_md_blocking_ast, 0);
523         ll_finish_md_op_data(op_data);
524         if (rc < 0)
525                 GOTO(out, retval = ERR_PTR(rc));
526
527         rc = ll_lookup_it_finish(req, it, &icbd);
528         if (rc != 0) {
529                 ll_intent_release(it);
530                 GOTO(out, retval = ERR_PTR(rc));
531         }
532
533         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
534             !S_ISREG(dentry->d_inode->i_mode) &&
535             !S_ISDIR(dentry->d_inode->i_mode)) {
536                 ll_release_openhandle(dentry, it);
537         }
538         ll_lookup_finish_locks(it, dentry);
539
540         if (dentry == save)
541                 GOTO(out, retval = NULL);
542         else
543                 GOTO(out, retval = dentry);
544  out:
545         if (req)
546                 ptlrpc_req_finished(req);
547         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
548                 ll_statahead_mark(parent, dentry);
549         return retval;
550 }
551
552 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
553                                         int lookup_flags)
554 {
555         struct lookup_intent *it;
556
557         OBD_ALLOC(it, sizeof(*it));
558         if (!it)
559                 return ERR_PTR(-ENOMEM);
560
561         if (lookup_flags & LOOKUP_OPEN) {
562                 it->it_op = IT_OPEN;
563                 if (lookup_flags & LOOKUP_CREATE)
564                         it->it_op |= IT_CREAT;
565                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
566                 it->it_flags = oit->flags;
567         } else {
568                 it->it_op = IT_GETATTR;
569         }
570
571         return it;
572 }
573
574 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
575                                    struct nameidata *nd)
576 {
577         struct dentry *de;
578         ENTRY;
579
580         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
581                 struct lookup_intent *it;
582
583 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17)
584                 /* Did we came here from failed revalidate just to propagate
585                  * its error? */
586                 if (nd->flags & LOOKUP_OPEN)
587                         if (IS_ERR(nd->intent.open.file))
588                                 RETURN((struct dentry *)nd->intent.open.file);
589 #endif
590                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
591                         it = ll_d2d(dentry)->lld_it;
592                         ll_d2d(dentry)->lld_it = NULL;
593                 } else {
594                         if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
595                                 /* We are sure this is new dentry, so we need to create
596                                    our private data and set the dentry ops */ 
597                                 ll_dops_init(dentry, 1, 1);
598                                 __d_lustre_invalidate(dentry);
599                                 d_add(dentry, NULL);
600                                 RETURN(NULL);
601                         }
602                         it = ll_convert_intent(&nd->intent.open, nd->flags);
603                         if (IS_ERR(it))
604                                 RETURN((struct dentry *)it);
605                 }
606
607                 de = ll_lookup_it(parent, dentry, it, nd->flags);
608                 if (de)
609                         dentry = de;
610                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
611                         if (dentry->d_inode &&
612                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
613                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
614                                         // We cannot call open here as it would
615                                         // deadlock.
616                                         ptlrpc_req_finished(
617                                                        (struct ptlrpc_request *)
618                                                           it->d.lustre.it_data);
619                                 } else {
620                                         struct file *filp;
621
622                                         nd->intent.open.file->private_data = it;
623                                         filp = lookup_instantiate_filp(nd,
624                                                                        dentry,
625                                                                        NULL);
626                                         if (IS_ERR(filp)) {
627                                                 if (de)
628                                                         dput(de);
629                                                 de = (struct dentry *)filp;
630                                         }
631                                 }
632                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
633                                 // XXX This can only reliably work on assumption
634                                 // that there are NO hashed negative dentries.
635                                 ll_d2d(dentry)->lld_it = it;
636                                 it = NULL; /* Will be freed in ll_create_nd */
637                                 /* We absolutely depend on ll_create_nd to be
638                                  * called to not leak this intent and possible
639                                  * data attached to it */
640                         }
641                 }
642
643                 if (it) {
644                         ll_intent_release(it);
645                         OBD_FREE(it, sizeof(*it));
646                 }
647         } else {
648                 de = ll_lookup_it(parent, dentry, NULL, 0);
649         }
650
651         RETURN(de);
652 }
653
654 /* We depend on "mode" being set with the proper file type/umask by now */
655 static struct inode *ll_create_node(struct inode *dir, const char *name,
656                                     int namelen, const void *data, int datalen,
657                                     int mode, __u64 extra,
658                                     struct lookup_intent *it)
659 {
660         struct inode *inode = NULL;
661         struct ptlrpc_request *request = NULL;
662         struct ll_sb_info *sbi = ll_i2sbi(dir);
663         int rc;
664         ENTRY;
665
666         LASSERT(it && it->d.lustre.it_disposition);
667
668         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
669         request = it->d.lustre.it_data;
670         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
671         rc = ll_prep_inode(&inode, request, dir->i_sb);
672         if (rc)
673                 GOTO(out, inode = ERR_PTR(rc));
674
675         LASSERT(list_empty(&inode->i_dentry));
676
677         /* We asked for a lock on the directory, but were granted a
678          * lock on the inode.  Since we finally have an inode pointer,
679          * stuff it in the lock. */
680         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
681                inode, inode->i_ino, inode->i_generation);
682         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
683         EXIT;
684  out:
685         ptlrpc_req_finished(request);
686         return inode;
687 }
688
689 /*
690  * By the time this is called, we already have created the directory cache
691  * entry for the new file, but it is so far negative - it has no inode.
692  *
693  * We defer creating the OBD object(s) until open, to keep the intent and
694  * non-intent code paths similar, and also because we do not have the MDS
695  * inode number before calling ll_create_node() (which is needed for LOV),
696  * so we would need to do yet another RPC to the MDS to store the LOV EA
697  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
698  * lmm_size in datalen (the MDS still has code which will handle that).
699  *
700  * If the create succeeds, we fill in the inode information
701  * with d_instantiate().
702  */
703 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
704                         struct lookup_intent *it)
705 {
706         struct inode *inode;
707         int rc = 0;
708         ENTRY;
709
710         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
711                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
712                dir->i_generation, dir, LL_IT2STR(it));
713
714         rc = it_open_error(DISP_OPEN_CREATE, it);
715         if (rc)
716                 RETURN(rc);
717
718         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
719                                NULL, 0, mode, 0, it);
720         if (IS_ERR(inode))
721                 RETURN(PTR_ERR(inode));
722
723         d_instantiate(dentry, inode);
724         RETURN(0);
725 }
726
727 static void ll_update_times(struct ptlrpc_request *request,
728                             struct inode *inode)
729 {
730         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
731                                                        &RMF_MDT_BODY);
732
733         LASSERT(body);
734         if (body->valid & OBD_MD_FLMTIME &&
735             body->mtime > LTIME_S(inode->i_mtime)) {
736                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
737                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
738                 LTIME_S(inode->i_mtime) = body->mtime;
739         }
740         if (body->valid & OBD_MD_FLCTIME &&
741             body->ctime > LTIME_S(inode->i_ctime))
742                 LTIME_S(inode->i_ctime) = body->ctime;
743 }
744
745 static int ll_new_node(struct inode *dir, struct qstr *name,
746                        const char *tgt, int mode, int rdev,
747                        struct dentry *dchild, __u32 opc)
748 {
749         struct ptlrpc_request *request = NULL;
750         struct md_op_data *op_data;
751         struct inode *inode = NULL;
752         struct ll_sb_info *sbi = ll_i2sbi(dir);
753         int tgt_len = 0;
754         int err;
755
756         ENTRY;
757         if (unlikely(tgt != NULL))
758                 tgt_len = strlen(tgt) + 1;
759
760         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
761                                      name->len, 0, opc, NULL);
762         if (IS_ERR(op_data))
763                 GOTO(err_exit, err = PTR_ERR(op_data));
764
765         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
766                         cfs_curproc_fsuid(), cfs_curproc_fsgid(),
767                         cfs_curproc_cap_pack(), rdev, &request);
768         ll_finish_md_op_data(op_data);
769         if (err)
770                 GOTO(err_exit, err);
771
772         ll_update_times(request, dir);
773
774         if (dchild) {
775                 err = ll_prep_inode(&inode, request, dchild->d_sb);
776                 if (err)
777                      GOTO(err_exit, err);
778
779                 d_instantiate(dchild, inode);
780         }
781         EXIT;
782 err_exit:
783         ptlrpc_req_finished(request);
784
785         return err;
786 }
787
788 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
789                             unsigned rdev, struct dentry *dchild)
790 {
791         int err;
792         ENTRY;
793
794         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
795                name->len, name->name, dir->i_ino, dir->i_generation, dir,
796                mode, rdev);
797
798         mode &= ~cfs_curproc_umask();
799
800         switch (mode & S_IFMT) {
801         case 0:
802                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
803         case S_IFREG:
804         case S_IFCHR:
805         case S_IFBLK:
806         case S_IFIFO:
807         case S_IFSOCK:
808                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
809                                   LUSTRE_OPC_MKNOD);
810                 break;
811         case S_IFDIR:
812                 err = -EPERM;
813                 break;
814         default:
815                 err = -EINVAL;
816         }
817
818         if (!err)
819                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
820
821         RETURN(err);
822 }
823
824 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
825                         int mode, struct nameidata *nd)
826 {
827         struct lookup_intent *it = ll_d2d(dentry)->lld_it;
828         int rc;
829
830         if (!it)
831                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
832
833         ll_d2d(dentry)->lld_it = NULL;
834
835         /* Was there an error? Propagate it! */
836         if (it->d.lustre.it_status) {
837                 rc = it->d.lustre.it_status;
838                 goto out;
839         }
840
841         rc = ll_create_it(dir, dentry, mode, it);
842         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
843                 struct file *filp;
844
845                 nd->intent.open.file->private_data = it;
846                 filp = lookup_instantiate_filp(nd, dentry, NULL);
847                 if (IS_ERR(filp))
848                         rc = PTR_ERR(filp);
849         }
850
851 out:
852         ll_intent_release(it);
853         OBD_FREE(it, sizeof(*it));
854
855         if (!rc)
856                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
857
858         return rc;
859 }
860
861 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
862                               const char *tgt, struct dentry *dchild)
863 {
864         int err;
865         ENTRY;
866
867         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
868                name->len, name->name, dir->i_ino, dir->i_generation,
869                dir, 3000, tgt);
870
871         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
872                           0, dchild, LUSTRE_OPC_SYMLINK);
873
874         if (!err)
875                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
876
877         RETURN(err);
878 }
879
880 static int ll_link_generic(struct inode *src,  struct inode *dir,
881                            struct qstr *name, struct dentry *dchild)
882 {
883         struct ll_sb_info *sbi = ll_i2sbi(dir);
884         struct ptlrpc_request *request = NULL;
885         struct md_op_data *op_data;
886         int err;
887
888         ENTRY;
889         CDEBUG(D_VFSTRACE,
890                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
891                src->i_ino, src->i_generation, src, dir->i_ino,
892                dir->i_generation, dir, name->len, name->name);
893
894         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
895                                      0, LUSTRE_OPC_ANY, NULL);
896         if (IS_ERR(op_data))
897                 RETURN(PTR_ERR(op_data));
898
899         err = md_link(sbi->ll_md_exp, op_data, &request);
900         ll_finish_md_op_data(op_data);
901         if (err)
902                 GOTO(out, err);
903
904         ll_update_times(request, dir);
905         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
906         EXIT;
907 out:
908         ptlrpc_req_finished(request);
909         RETURN(err);
910 }
911
912 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
913                             int mode, struct dentry *dchild)
914
915 {
916         int err;
917         ENTRY;
918
919         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
920                name->len, name->name, dir->i_ino, dir->i_generation, dir);
921
922         mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
923         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
924
925         if (!err)
926                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
927
928         RETURN(err);
929 }
930
931 /* Try to find the child dentry by its name.
932    If found, put the result fid into @fid. */
933 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
934                              struct lu_fid *fid)
935 {
936         struct dentry *parent, *child;
937
938         parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
939         child = d_lookup(parent, name);
940         if (child) {
941                 if (child->d_inode)
942                         *fid = *ll_inode2fid(child->d_inode);
943                 dput(child);
944         }
945 }
946
947 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
948                             struct dentry *dchild, struct qstr *name)
949 {
950         struct ptlrpc_request *request = NULL;
951         struct md_op_data *op_data;
952         int rc;
953         ENTRY;
954
955         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
956                name->len, name->name, dir->i_ino, dir->i_generation, dir);
957
958         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
959                 RETURN(-EBUSY);
960
961         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
962                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
963         if (IS_ERR(op_data))
964                 RETURN(PTR_ERR(op_data));
965
966         ll_get_child_fid(dir, name, &op_data->op_fid3);
967         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
968         ll_finish_md_op_data(op_data);
969         if (rc == 0) {
970                 ll_update_times(request, dir);
971                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
972         }
973
974         ptlrpc_req_finished(request);
975         RETURN(rc);
976 }
977
978 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
979 {
980         struct mdt_body *body;
981         struct lov_mds_md *eadata;
982         struct lov_stripe_md *lsm = NULL;
983         struct obd_trans_info oti = { 0 };
984         struct obdo *oa;
985         struct obd_capa *oc = NULL;
986         int rc;
987         ENTRY;
988
989         /* req is swabbed so this is safe */
990         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
991         if (!(body->valid & OBD_MD_FLEASIZE))
992                 RETURN(0);
993
994         if (body->eadatasize == 0) {
995                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
996                 GOTO(out, rc = -EPROTO);
997         }
998
999         /* The MDS sent back the EA because we unlinked the last reference
1000          * to this file. Use this EA to unlink the objects on the OST.
1001          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1002          * check it is complete and sensible. */
1003         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1004                                               body->eadatasize);
1005         LASSERT(eadata != NULL);
1006
1007         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1008         if (rc < 0) {
1009                 CERROR("obd_unpackmd: %d\n", rc);
1010                 GOTO(out, rc);
1011         }
1012         LASSERT(rc >= sizeof(*lsm));
1013
1014         OBDO_ALLOC(oa);
1015         if (oa == NULL)
1016                 GOTO(out_free_memmd, rc = -ENOMEM);
1017
1018         oa->o_id = lsm->lsm_object_id;
1019         oa->o_seq = lsm->lsm_object_seq;
1020         oa->o_mode = body->mode & S_IFMT;
1021         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1022
1023         if (body->valid & OBD_MD_FLCOOKIE) {
1024                 oa->o_valid |= OBD_MD_FLCOOKIE;
1025                 oti.oti_logcookies =
1026                         req_capsule_server_sized_get(&request->rq_pill,
1027                                                      &RMF_LOGCOOKIES,
1028                                                    sizeof(struct llog_cookie) *
1029                                                      lsm->lsm_stripe_count);
1030                 if (oti.oti_logcookies == NULL) {
1031                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1032                         body->valid &= ~OBD_MD_FLCOOKIE;
1033                 }
1034         }
1035
1036         if (body->valid & OBD_MD_FLOSSCAPA) {
1037                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1038                 if (rc)
1039                         GOTO(out_free_memmd, rc);
1040         }
1041
1042         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1043                          ll_i2mdexp(dir), oc);
1044         capa_put(oc);
1045         OBDO_FREE(oa);
1046         if (rc)
1047                 CERROR("obd destroy objid "LPX64" error %d\n",
1048                        lsm->lsm_object_id, rc);
1049  out_free_memmd:
1050         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1051  out:
1052         return rc;
1053 }
1054
1055 /* ll_unlink_generic() doesn't update the inode with the new link count.
1056  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1057  * is any lock existing. They will recycle dentries and inodes based upon locks
1058  * too. b=20433 */
1059 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1060                              struct dentry *dchild, struct qstr *name)
1061 {
1062         struct ptlrpc_request *request = NULL;
1063         struct md_op_data *op_data;
1064         int rc;
1065         ENTRY;
1066         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1067                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1068
1069         /*
1070          * XXX: unlink bind mountpoint maybe call to here,
1071          * just check it as vfs_unlink does.
1072          */
1073         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1074                 RETURN(-EBUSY);
1075
1076         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1077                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1078         if (IS_ERR(op_data))
1079                 RETURN(PTR_ERR(op_data));
1080
1081         ll_get_child_fid(dir, name, &op_data->op_fid3);
1082         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1083         ll_finish_md_op_data(op_data);
1084         if (rc)
1085                 GOTO(out, rc);
1086
1087         ll_update_times(request, dir);
1088         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1089
1090         rc = ll_objects_destroy(request, dir);
1091  out:
1092         ptlrpc_req_finished(request);
1093         RETURN(rc);
1094 }
1095
1096 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1097                              struct dentry *src_dchild, struct qstr *src_name,
1098                              struct inode *tgt, struct dentry *tgt_dparent,
1099                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1100 {
1101         struct ptlrpc_request *request = NULL;
1102         struct ll_sb_info *sbi = ll_i2sbi(src);
1103         struct md_op_data *op_data;
1104         int err;
1105         ENTRY;
1106         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1107                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1108                src->i_ino, src->i_generation, src, tgt_name->len,
1109                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1110
1111         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1112             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1113                 RETURN(-EBUSY);
1114
1115         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1116                                      LUSTRE_OPC_ANY, NULL);
1117         if (IS_ERR(op_data))
1118                 RETURN(PTR_ERR(op_data));
1119
1120         ll_get_child_fid(src, src_name, &op_data->op_fid3);
1121         ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1122         err = md_rename(sbi->ll_md_exp, op_data,
1123                         src_name->name, src_name->len,
1124                         tgt_name->name, tgt_name->len, &request);
1125         ll_finish_md_op_data(op_data);
1126         if (!err) {
1127                 ll_update_times(request, src);
1128                 ll_update_times(request, tgt);
1129                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1130                 err = ll_objects_destroy(request, src);
1131         }
1132
1133         ptlrpc_req_finished(request);
1134
1135         RETURN(err);
1136 }
1137
1138 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1139                     ll_dev_t rdev)
1140 {
1141         return ll_mknod_generic(dir, &dchild->d_name, mode,
1142                                 old_encode_dev(rdev), dchild);
1143 }
1144
1145 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1146 {
1147         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1148 }
1149 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1150 {
1151         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1152 }
1153 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1154 {
1155         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1156 }
1157 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1158                       const char *oldname)
1159 {
1160         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1161 }
1162 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1163                    struct dentry *new_dentry)
1164 {
1165         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1166                                new_dentry);
1167 }
1168 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1169                      struct inode *new_dir, struct dentry *new_dentry)
1170 {
1171         int err;
1172         err = ll_rename_generic(old_dir, NULL,
1173                                  old_dentry, &old_dentry->d_name,
1174                                  new_dir, NULL, new_dentry,
1175                                  &new_dentry->d_name);
1176         if (!err) {
1177 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1178                 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1179 #endif
1180                         d_move(old_dentry, new_dentry);
1181         }
1182         return err;
1183 }
1184
1185 struct inode_operations ll_dir_inode_operations = {
1186         .mknod              = ll_mknod,
1187         .lookup             = ll_lookup_nd,
1188         .create             = ll_create_nd,
1189         /* We need all these non-raw things for NFSD, to not patch it. */
1190         .unlink             = ll_unlink,
1191         .mkdir              = ll_mkdir,
1192         .rmdir              = ll_rmdir,
1193         .symlink            = ll_symlink,
1194         .link               = ll_link,
1195         .rename             = ll_rename,
1196         .setattr            = ll_setattr,
1197         .getattr            = ll_getattr,
1198         .permission         = ll_inode_permission,
1199         .setxattr           = ll_setxattr,
1200         .getxattr           = ll_getxattr,
1201         .listxattr          = ll_listxattr,
1202         .removexattr        = ll_removexattr,
1203 };
1204
1205 struct inode_operations ll_special_inode_operations = {
1206         .setattr        = ll_setattr,
1207         .getattr        = ll_getattr,
1208         .permission     = ll_inode_permission,
1209         .setxattr       = ll_setxattr,
1210         .getxattr       = ll_getxattr,
1211         .listxattr      = ll_listxattr,
1212         .removexattr    = ll_removexattr,
1213 };