Whamcloud - gitweb
331db4e94baddbc9f23057ceee8f69707c556316
[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/smp_lock.h>
41 #include <linux/quotaops.h>
42 #include <linux/highmem.h>
43 #include <linux/pagemap.h>
44 #include <linux/security.h>
45
46 #define DEBUG_SUBSYSTEM S_LLITE
47
48 #include <obd_support.h>
49 #include <lustre_fid.h>
50 #include <lustre_lite.h>
51 #include <lustre_dlm.h>
52 #include <lustre_ver.h>
53 #include "llite_internal.h"
54
55 /*
56  * Check if we have something mounted at the named dchild.
57  * In such a case there would always be dentry present.
58  */
59 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
60                            struct qstr *name)
61 {
62         int mounted = 0;
63
64         if (unlikely(dchild)) {
65                 mounted = d_mountpoint(dchild);
66         } else if (dparent) {
67                 dchild = d_lookup(dparent, name);
68                 if (dchild) {
69                         mounted = d_mountpoint(dchild);
70                         dput(dchild);
71                 }
72         }
73         return mounted;
74 }
75
76 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
77 {
78         ENTRY;
79
80         ldlm_lock_decref(lockh, mode);
81
82         RETURN(0);
83 }
84
85
86 /* called from iget5_locked->find_inode() under inode_lock spinlock */
87 static int ll_test_inode(struct inode *inode, void *opaque)
88 {
89         struct ll_inode_info *lli = ll_i2info(inode);
90         struct lustre_md     *md = opaque;
91
92         if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
93                 CERROR("MDS body missing FID\n");
94                 return 0;
95         }
96
97         if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
98                 return 0;
99
100         return 1;
101 }
102
103 static int ll_set_inode(struct inode *inode, void *opaque)
104 {
105         struct ll_inode_info *lli = ll_i2info(inode);
106         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
107
108         if (unlikely(!(body->valid & OBD_MD_FLID))) {
109                 CERROR("MDS body missing FID\n");
110                 return -EINVAL;
111         }
112
113         lli->lli_fid = body->fid1;
114         if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
115                 CERROR("Can not initialize inode "DFID" without object type: "
116                        "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
117                 return -EINVAL;
118         }
119
120         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
121         if (unlikely(inode->i_mode == 0)) {
122                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
123                 return -EINVAL;
124         }
125
126         ll_lli_init(lli);
127
128         return 0;
129 }
130
131
132 /*
133  * Get an inode by inode number (already instantiated by the intent lookup).
134  * Returns inode or NULL
135  */
136 struct inode *ll_iget(struct super_block *sb, ino_t hash,
137                       struct lustre_md *md)
138 {
139         struct inode         *inode;
140         ENTRY;
141
142         LASSERT(hash != 0);
143         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
144
145         if (inode) {
146                 if (inode->i_state & I_NEW) {
147                         int rc = 0;
148
149                         ll_read_inode2(inode, md);
150                         if (S_ISREG(inode->i_mode) &&
151                             ll_i2info(inode)->lli_clob == NULL)
152                                 rc = cl_file_inode_init(inode, md);
153                         if (rc != 0) {
154                                 md->lsm = NULL;
155                                 make_bad_inode(inode);
156                                 unlock_new_inode(inode);
157                                 iput(inode);
158                                 inode = ERR_PTR(rc);
159                         } else
160                                 unlock_new_inode(inode);
161                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
162                         ll_update_inode(inode, md);
163                 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
164                        inode, PFID(&md->body->fid1));
165         }
166         RETURN(inode);
167 }
168
169 static void ll_invalidate_negative_children(struct inode *dir)
170 {
171         struct dentry *dentry, *tmp_subdir;
172
173         ll_lock_dcache(dir);
174         list_for_each_entry(dentry, &dir->i_dentry, d_alias) {
175                 spin_lock(&dentry->d_lock);
176                 if (!list_empty(&dentry->d_subdirs)) {
177                         struct dentry *child;
178
179                         list_for_each_entry_safe(child, tmp_subdir,
180                                                  &dentry->d_subdirs,
181                                                  d_child) {
182                                 if (child->d_inode == NULL)
183                                         d_lustre_invalidate(child);
184                         }
185                 }
186                 spin_unlock(&dentry->d_lock);
187         }
188         ll_unlock_dcache(dir);
189 }
190
191 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
192                        void *data, int flag)
193 {
194         int rc;
195         struct lustre_handle lockh;
196         ENTRY;
197
198         switch (flag) {
199         case LDLM_CB_BLOCKING:
200                 ldlm_lock2handle(lock, &lockh);
201                 rc = ldlm_cli_cancel(&lockh);
202                 if (rc < 0) {
203                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
204                         RETURN(rc);
205                 }
206                 break;
207         case LDLM_CB_CANCELING: {
208                 struct inode *inode = ll_inode_from_lock(lock);
209                 struct ll_inode_info *lli;
210                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
211                 struct lu_fid *fid;
212                 ldlm_mode_t mode = lock->l_req_mode;
213
214                 /* Invalidate all dentries associated with this inode */
215                 if (inode == NULL)
216                         break;
217
218                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
219                 /* For OPEN locks we differentiate between lock modes - CR, CW. PR - bug 22891 */
220                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE))
221                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
222
223                 if (bits & MDS_INODELOCK_OPEN)
224                         ll_have_md_lock(inode, &bits, mode);
225
226                 fid = ll_inode2fid(inode);
227                 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
228                     lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
229                     lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
230                         LDLM_ERROR(lock, "data mismatch with object "
231                                    DFID" (%p)", PFID(fid), inode);
232                 }
233
234                 if (bits & MDS_INODELOCK_OPEN) {
235                         int flags = 0;
236                         switch (lock->l_req_mode) {
237                         case LCK_CW:
238                                 flags = FMODE_WRITE;
239                                 break;
240                         case LCK_PR:
241                                 flags = FMODE_EXEC;
242                                 break;
243                         case LCK_CR:
244                                 flags = FMODE_READ;
245                                 break;
246                         default:
247                                 CERROR("Unexpected lock mode for OPEN lock "
248                                        "%d, inode %ld\n", lock->l_req_mode,
249                                        inode->i_ino);
250                         }
251                         ll_md_real_close(inode, flags);
252                 }
253
254                 lli = ll_i2info(inode);
255                 if (bits & MDS_INODELOCK_UPDATE)
256                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
257
258                 if (S_ISDIR(inode->i_mode) &&
259                      (bits & MDS_INODELOCK_UPDATE)) {
260                         CDEBUG(D_INODE, "invalidating inode %lu\n",
261                                inode->i_ino);
262                         truncate_inode_pages(inode->i_mapping, 0);
263                         ll_invalidate_negative_children(inode);
264                 }
265
266                 if (inode->i_sb->s_root &&
267                     inode != inode->i_sb->s_root->d_inode &&
268                     (bits & MDS_INODELOCK_LOOKUP))
269                         ll_invalidate_aliases(inode);
270                 iput(inode);
271                 break;
272         }
273         default:
274                 LBUG();
275         }
276
277         RETURN(0);
278 }
279
280 __u32 ll_i2suppgid(struct inode *i)
281 {
282         if (cfs_curproc_is_in_groups(i->i_gid))
283                 return (__u32)i->i_gid;
284         else
285                 return (__u32)(-1);
286 }
287
288 /* Pack the required supplementary groups into the supplied groups array.
289  * If we don't need to use the groups from the target inode(s) then we
290  * instead pack one or more groups from the user's supplementary group
291  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
292 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
293 {
294 #if 0
295         int i;
296 #endif
297
298         LASSERT(i1 != NULL);
299         LASSERT(suppgids != NULL);
300
301         suppgids[0] = ll_i2suppgid(i1);
302
303         if (i2)
304                 suppgids[1] = ll_i2suppgid(i2);
305                 else
306                         suppgids[1] = -1;
307
308 #if 0
309         for (i = 0; i < current_ngroups; i++) {
310                 if (suppgids[0] == -1) {
311                         if (current_groups[i] != suppgids[1])
312                                 suppgids[0] = current_groups[i];
313                         continue;
314                 }
315                 if (suppgids[1] == -1) {
316                         if (current_groups[i] != suppgids[0])
317                                 suppgids[1] = current_groups[i];
318                         continue;
319                 }
320                 break;
321         }
322 #endif
323 }
324
325 /*
326  * try to reuse three types of dentry:
327  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
328  *    by concurrent .revalidate).
329  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
330  *    be cleared by others calling d_lustre_revalidate).
331  * 3. DISCONNECTED alias.
332  */
333 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
334 {
335         struct dentry *alias, *discon_alias, *invalid_alias;
336
337         if (list_empty(&inode->i_dentry))
338                 return NULL;
339
340         discon_alias = invalid_alias = NULL;
341
342         ll_lock_dcache(inode);
343         list_for_each_entry(alias, &inode->i_dentry, d_alias) {
344                 LASSERT(alias != dentry);
345
346                 spin_lock(&alias->d_lock);
347                 if (alias->d_flags & DCACHE_DISCONNECTED)
348                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
349                         discon_alias = alias;
350                 else if (alias->d_parent == dentry->d_parent             &&
351                          alias->d_name.hash == dentry->d_name.hash       &&
352                          alias->d_name.len == dentry->d_name.len         &&
353                          memcmp(alias->d_name.name, dentry->d_name.name,
354                                 dentry->d_name.len) == 0)
355                         invalid_alias = alias;
356                 spin_unlock(&alias->d_lock);
357
358                 if (invalid_alias)
359                         break;
360         }
361         alias = invalid_alias ?: discon_alias ?: NULL;
362         if (alias) {
363                 spin_lock(&alias->d_lock);
364                 dget_dlock(alias);
365                 spin_unlock(&alias->d_lock);
366         }
367         ll_unlock_dcache(inode);
368
369         return alias;
370 }
371
372 /*
373  * Similar to d_splice_alias(), but lustre treats DCACHE_LUSTRE_INVALID alias
374  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
375  */
376 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
377 {
378         struct dentry *new;
379
380         if (inode) {
381                 new = ll_find_alias(inode, de);
382                 if (new) {
383                         ll_dops_init(new, 1, 1);
384                         d_move(new, de);
385                         iput(inode);
386                         CDEBUG(D_DENTRY,
387                                "Reuse dentry %p inode %p refc %d flags %#x\n",
388                               new, new->d_inode, d_refcount(new), new->d_flags);
389                         return new;
390                 }
391         }
392         __d_lustre_invalidate(de);
393         ll_dops_init(de, 1, 1);
394         d_add(de, inode);
395         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
396                de, de->d_inode, d_refcount(de), de->d_flags);
397         return de;
398 }
399
400 int ll_lookup_it_finish(struct ptlrpc_request *request,
401                         struct lookup_intent *it, void *data)
402 {
403         struct it_cb_data *icbd = data;
404         struct dentry **de = icbd->icbd_childp;
405         struct inode *parent = icbd->icbd_parent;
406         struct inode *inode = NULL;
407         __u64 bits = 0;
408         int rc;
409         ENTRY;
410
411         /* NB 1 request reference will be taken away by ll_intent_lock()
412          * when I return */
413         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
414                 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
415                 if (rc)
416                         RETURN(rc);
417
418                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
419
420                 /* We used to query real size from OSTs here, but actually
421                    this is not needed. For stat() calls size would be updated
422                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
423                    2.4 and
424                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
425                    Everybody else who needs correct file size would call
426                    ll_glimpse_size or some equivalent themselves anyway.
427                    Also see bug 7198. */
428         }
429
430         *de = ll_splice_alias(inode, *de);
431
432         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
433                 /* we have lookup look - unhide dentry */
434                 if (bits & MDS_INODELOCK_LOOKUP)
435                         d_lustre_revalidate(*de);
436         } else {
437                 /* Check that parent has UPDATE lock. */
438                 struct lookup_intent parent_it = {
439                                         .it_op = IT_GETATTR,
440                                         .d.lustre.it_lock_handle = 0 };
441
442                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
443                                        &ll_i2info(parent)->lli_fid, NULL)) {
444                         d_lustre_revalidate(*de);
445                         ll_intent_release(&parent_it);
446                 }
447         }
448
449         RETURN(0);
450 }
451
452 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
453                                    struct lookup_intent *it, int lookup_flags)
454 {
455         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
456         struct dentry *save = dentry, *retval;
457         struct ptlrpc_request *req = NULL;
458         struct md_op_data *op_data;
459         struct it_cb_data icbd;
460         __u32 opc;
461         int rc;
462         ENTRY;
463
464         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
465                 RETURN(ERR_PTR(-ENAMETOOLONG));
466
467         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
468                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
469                parent->i_generation, parent, LL_IT2STR(it));
470
471         if (d_mountpoint(dentry))
472                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
473
474         ll_frob_intent(&it, &lookup_it);
475
476         /* As do_lookup is called before follow_mount, root dentry may be left
477          * not valid, revalidate it here. */
478         if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
479             (it->it_op & (IT_OPEN | IT_CREAT))) {
480                 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
481                                             MDS_INODELOCK_LOOKUP);
482                 if (rc)
483                         RETURN(ERR_PTR(rc));
484         }
485
486         if (it->it_op == IT_GETATTR) {
487                 rc = ll_statahead_enter(parent, &dentry, 0);
488                 if (rc == 1) {
489                         if (dentry == save)
490                                 GOTO(out, retval = NULL);
491                         GOTO(out, retval = dentry);
492                 }
493         }
494
495         icbd.icbd_childp = &dentry;
496         icbd.icbd_parent = parent;
497
498         if (it->it_op & IT_CREAT ||
499             (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
500                 opc = LUSTRE_OPC_CREATE;
501         else
502                 opc = LUSTRE_OPC_ANY;
503
504         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
505                                      dentry->d_name.len, lookup_flags, opc,
506                                      NULL);
507         if (IS_ERR(op_data))
508                 RETURN((void *)op_data);
509
510         it->it_create_mode &= ~cfs_curproc_umask();
511
512         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
513                             lookup_flags, &req, ll_md_blocking_ast, 0);
514         ll_finish_md_op_data(op_data);
515         if (rc < 0)
516                 GOTO(out, retval = ERR_PTR(rc));
517
518         rc = ll_lookup_it_finish(req, it, &icbd);
519         if (rc != 0) {
520                 ll_intent_release(it);
521                 GOTO(out, retval = ERR_PTR(rc));
522         }
523
524         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
525             !S_ISREG(dentry->d_inode->i_mode) &&
526             !S_ISDIR(dentry->d_inode->i_mode)) {
527                 ll_release_openhandle(dentry, it);
528         }
529         ll_lookup_finish_locks(it, dentry);
530
531         if (dentry == save)
532                 GOTO(out, retval = NULL);
533         else
534                 GOTO(out, retval = dentry);
535  out:
536         if (req)
537                 ptlrpc_req_finished(req);
538         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
539                 ll_statahead_mark(parent, dentry);
540         return retval;
541 }
542
543 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
544                                         int lookup_flags)
545 {
546         struct lookup_intent *it;
547
548         OBD_ALLOC(it, sizeof(*it));
549         if (!it)
550                 return ERR_PTR(-ENOMEM);
551
552         if (lookup_flags & LOOKUP_OPEN) {
553                 it->it_op = IT_OPEN;
554                 if (lookup_flags & LOOKUP_CREATE)
555                         it->it_op |= IT_CREAT;
556                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
557                 it->it_flags = oit->flags;
558         } else {
559                 it->it_op = IT_GETATTR;
560         }
561
562         return it;
563 }
564
565 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
566                                    struct nameidata *nd)
567 {
568         struct dentry *de;
569         ENTRY;
570
571         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
572                 struct lookup_intent *it;
573
574 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17)
575                 /* Did we came here from failed revalidate just to propagate
576                  * its error? */
577                 if (nd->flags & LOOKUP_OPEN)
578                         if (IS_ERR(nd->intent.open.file))
579                                 RETURN((struct dentry *)nd->intent.open.file);
580 #endif
581                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
582                         it = ll_d2d(dentry)->lld_it;
583                         ll_d2d(dentry)->lld_it = NULL;
584                 } else {
585                         if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
586                                 /* We are sure this is new dentry, so we need to create
587                                    our private data and set the dentry ops */ 
588                                 ll_dops_init(dentry, 1, 1);
589                                 __d_lustre_invalidate(dentry);
590                                 d_add(dentry, NULL);
591                                 RETURN(NULL);
592                         }
593                         it = ll_convert_intent(&nd->intent.open, nd->flags);
594                         if (IS_ERR(it))
595                                 RETURN((struct dentry *)it);
596                 }
597
598                 de = ll_lookup_it(parent, dentry, it, nd->flags);
599                 if (de)
600                         dentry = de;
601                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
602                         if (dentry->d_inode &&
603                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
604                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
605                                         // We cannot call open here as it would
606                                         // deadlock.
607                                         ptlrpc_req_finished(
608                                                        (struct ptlrpc_request *)
609                                                           it->d.lustre.it_data);
610                                 } else {
611                                         struct file *filp;
612
613                                         nd->intent.open.file->private_data = it;
614                                         filp = lookup_instantiate_filp(nd,
615                                                                        dentry,
616                                                                        NULL);
617                                         if (IS_ERR(filp)) {
618                                                 if (de)
619                                                         dput(de);
620                                                 de = (struct dentry *)filp;
621                                         }
622                                 }
623                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
624                                 // XXX This can only reliably work on assumption
625                                 // that there are NO hashed negative dentries.
626                                 ll_d2d(dentry)->lld_it = it;
627                                 it = NULL; /* Will be freed in ll_create_nd */
628                                 /* We absolutely depend on ll_create_nd to be
629                                  * called to not leak this intent and possible
630                                  * data attached to it */
631                         }
632                 }
633
634                 if (it) {
635                         ll_intent_release(it);
636                         OBD_FREE(it, sizeof(*it));
637                 }
638         } else {
639                 de = ll_lookup_it(parent, dentry, NULL, 0);
640         }
641
642         RETURN(de);
643 }
644
645 /* We depend on "mode" being set with the proper file type/umask by now */
646 static struct inode *ll_create_node(struct inode *dir, const char *name,
647                                     int namelen, const void *data, int datalen,
648                                     int mode, __u64 extra,
649                                     struct lookup_intent *it)
650 {
651         struct inode *inode = NULL;
652         struct ptlrpc_request *request = NULL;
653         struct ll_sb_info *sbi = ll_i2sbi(dir);
654         int rc;
655         ENTRY;
656
657         LASSERT(it && it->d.lustre.it_disposition);
658
659         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
660         request = it->d.lustre.it_data;
661         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
662         rc = ll_prep_inode(&inode, request, dir->i_sb);
663         if (rc)
664                 GOTO(out, inode = ERR_PTR(rc));
665
666         LASSERT(list_empty(&inode->i_dentry));
667
668         /* We asked for a lock on the directory, but were granted a
669          * lock on the inode.  Since we finally have an inode pointer,
670          * stuff it in the lock. */
671         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
672                inode, inode->i_ino, inode->i_generation);
673         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
674         EXIT;
675  out:
676         ptlrpc_req_finished(request);
677         return inode;
678 }
679
680 /*
681  * By the time this is called, we already have created the directory cache
682  * entry for the new file, but it is so far negative - it has no inode.
683  *
684  * We defer creating the OBD object(s) until open, to keep the intent and
685  * non-intent code paths similar, and also because we do not have the MDS
686  * inode number before calling ll_create_node() (which is needed for LOV),
687  * so we would need to do yet another RPC to the MDS to store the LOV EA
688  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
689  * lmm_size in datalen (the MDS still has code which will handle that).
690  *
691  * If the create succeeds, we fill in the inode information
692  * with d_instantiate().
693  */
694 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
695                         struct lookup_intent *it)
696 {
697         struct inode *inode;
698         int rc = 0;
699         ENTRY;
700
701         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
702                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
703                dir->i_generation, dir, LL_IT2STR(it));
704
705         rc = it_open_error(DISP_OPEN_CREATE, it);
706         if (rc)
707                 RETURN(rc);
708
709         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
710                                NULL, 0, mode, 0, it);
711         if (IS_ERR(inode))
712                 RETURN(PTR_ERR(inode));
713
714         d_instantiate(dentry, inode);
715         RETURN(0);
716 }
717
718 static void ll_update_times(struct ptlrpc_request *request,
719                             struct inode *inode)
720 {
721         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
722                                                        &RMF_MDT_BODY);
723
724         LASSERT(body);
725         if (body->valid & OBD_MD_FLMTIME &&
726             body->mtime > LTIME_S(inode->i_mtime)) {
727                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
728                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
729                 LTIME_S(inode->i_mtime) = body->mtime;
730         }
731         if (body->valid & OBD_MD_FLCTIME &&
732             body->ctime > LTIME_S(inode->i_ctime))
733                 LTIME_S(inode->i_ctime) = body->ctime;
734 }
735
736 static int ll_new_node(struct inode *dir, struct qstr *name,
737                        const char *tgt, int mode, int rdev,
738                        struct dentry *dchild, __u32 opc)
739 {
740         struct ptlrpc_request *request = NULL;
741         struct md_op_data *op_data;
742         struct inode *inode = NULL;
743         struct ll_sb_info *sbi = ll_i2sbi(dir);
744         int tgt_len = 0;
745         int err;
746
747         ENTRY;
748         if (unlikely(tgt != NULL))
749                 tgt_len = strlen(tgt) + 1;
750
751         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
752                                      name->len, 0, opc, NULL);
753         if (IS_ERR(op_data))
754                 GOTO(err_exit, err = PTR_ERR(op_data));
755
756         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
757                         cfs_curproc_fsuid(), cfs_curproc_fsgid(),
758                         cfs_curproc_cap_pack(), rdev, &request);
759         ll_finish_md_op_data(op_data);
760         if (err)
761                 GOTO(err_exit, err);
762
763         ll_update_times(request, dir);
764
765         if (dchild) {
766                 err = ll_prep_inode(&inode, request, dchild->d_sb);
767                 if (err)
768                      GOTO(err_exit, err);
769
770                 d_instantiate(dchild, inode);
771         }
772         EXIT;
773 err_exit:
774         ptlrpc_req_finished(request);
775
776         return err;
777 }
778
779 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
780                             unsigned rdev, struct dentry *dchild)
781 {
782         int err;
783         ENTRY;
784
785         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
786                name->len, name->name, dir->i_ino, dir->i_generation, dir,
787                mode, rdev);
788
789         mode &= ~cfs_curproc_umask();
790
791         switch (mode & S_IFMT) {
792         case 0:
793                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
794         case S_IFREG:
795         case S_IFCHR:
796         case S_IFBLK:
797         case S_IFIFO:
798         case S_IFSOCK:
799                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
800                                   LUSTRE_OPC_MKNOD);
801                 break;
802         case S_IFDIR:
803                 err = -EPERM;
804                 break;
805         default:
806                 err = -EINVAL;
807         }
808
809         if (!err)
810                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
811
812         RETURN(err);
813 }
814
815 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
816                         int mode, struct nameidata *nd)
817 {
818         struct lookup_intent *it = ll_d2d(dentry)->lld_it;
819         int rc;
820
821         if (!it)
822                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
823
824         ll_d2d(dentry)->lld_it = NULL;
825
826         /* Was there an error? Propagate it! */
827         if (it->d.lustre.it_status) {
828                 rc = it->d.lustre.it_status;
829                 goto out;
830         }
831
832         rc = ll_create_it(dir, dentry, mode, it);
833         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
834                 struct file *filp;
835
836                 nd->intent.open.file->private_data = it;
837                 filp = lookup_instantiate_filp(nd, dentry, NULL);
838                 if (IS_ERR(filp))
839                         rc = PTR_ERR(filp);
840         }
841
842 out:
843         ll_intent_release(it);
844         OBD_FREE(it, sizeof(*it));
845
846         if (!rc)
847                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
848
849         return rc;
850 }
851
852 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
853                               const char *tgt, struct dentry *dchild)
854 {
855         int err;
856         ENTRY;
857
858         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
859                name->len, name->name, dir->i_ino, dir->i_generation,
860                dir, 3000, tgt);
861
862         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
863                           0, dchild, LUSTRE_OPC_SYMLINK);
864
865         if (!err)
866                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
867
868         RETURN(err);
869 }
870
871 static int ll_link_generic(struct inode *src,  struct inode *dir,
872                            struct qstr *name, struct dentry *dchild)
873 {
874         struct ll_sb_info *sbi = ll_i2sbi(dir);
875         struct ptlrpc_request *request = NULL;
876         struct md_op_data *op_data;
877         int err;
878
879         ENTRY;
880         CDEBUG(D_VFSTRACE,
881                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
882                src->i_ino, src->i_generation, src, dir->i_ino,
883                dir->i_generation, dir, name->len, name->name);
884
885         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
886                                      0, LUSTRE_OPC_ANY, NULL);
887         if (IS_ERR(op_data))
888                 RETURN(PTR_ERR(op_data));
889
890         err = md_link(sbi->ll_md_exp, op_data, &request);
891         ll_finish_md_op_data(op_data);
892         if (err)
893                 GOTO(out, err);
894
895         ll_update_times(request, dir);
896         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
897         EXIT;
898 out:
899         ptlrpc_req_finished(request);
900         RETURN(err);
901 }
902
903 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
904                             int mode, struct dentry *dchild)
905
906 {
907         int err;
908         ENTRY;
909
910         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
911                name->len, name->name, dir->i_ino, dir->i_generation, dir);
912
913         mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
914         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
915
916         if (!err)
917                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
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, dir);
962                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
963         }
964
965         ptlrpc_req_finished(request);
966         RETURN(rc);
967 }
968
969 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
970 {
971         struct mdt_body *body;
972         struct lov_mds_md *eadata;
973         struct lov_stripe_md *lsm = NULL;
974         struct obd_trans_info oti = { 0 };
975         struct obdo *oa;
976         struct obd_capa *oc = NULL;
977         int rc;
978         ENTRY;
979
980         /* req is swabbed so this is safe */
981         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
982         if (!(body->valid & OBD_MD_FLEASIZE))
983                 RETURN(0);
984
985         if (body->eadatasize == 0) {
986                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
987                 GOTO(out, rc = -EPROTO);
988         }
989
990         /* The MDS sent back the EA because we unlinked the last reference
991          * to this file. Use this EA to unlink the objects on the OST.
992          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
993          * check it is complete and sensible. */
994         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
995                                               body->eadatasize);
996         LASSERT(eadata != NULL);
997
998         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
999         if (rc < 0) {
1000                 CERROR("obd_unpackmd: %d\n", rc);
1001                 GOTO(out, rc);
1002         }
1003         LASSERT(rc >= sizeof(*lsm));
1004
1005         OBDO_ALLOC(oa);
1006         if (oa == NULL)
1007                 GOTO(out_free_memmd, rc = -ENOMEM);
1008
1009         oa->o_id = lsm->lsm_object_id;
1010         oa->o_seq = lsm->lsm_object_seq;
1011         oa->o_mode = body->mode & S_IFMT;
1012         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1013
1014         if (body->valid & OBD_MD_FLCOOKIE) {
1015                 oa->o_valid |= OBD_MD_FLCOOKIE;
1016                 oti.oti_logcookies =
1017                         req_capsule_server_sized_get(&request->rq_pill,
1018                                                      &RMF_LOGCOOKIES,
1019                                                    sizeof(struct llog_cookie) *
1020                                                      lsm->lsm_stripe_count);
1021                 if (oti.oti_logcookies == NULL) {
1022                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1023                         body->valid &= ~OBD_MD_FLCOOKIE;
1024                 }
1025         }
1026
1027         if (body->valid & OBD_MD_FLOSSCAPA) {
1028                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1029                 if (rc)
1030                         GOTO(out_free_memmd, rc);
1031         }
1032
1033         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1034                          ll_i2mdexp(dir), oc);
1035         capa_put(oc);
1036         OBDO_FREE(oa);
1037         if (rc)
1038                 CERROR("obd destroy objid "LPX64" error %d\n",
1039                        lsm->lsm_object_id, rc);
1040  out_free_memmd:
1041         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1042  out:
1043         return rc;
1044 }
1045
1046 /* ll_unlink_generic() doesn't update the inode with the new link count.
1047  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1048  * is any lock existing. They will recycle dentries and inodes based upon locks
1049  * too. b=20433 */
1050 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1051                              struct dentry *dchild, struct qstr *name)
1052 {
1053         struct ptlrpc_request *request = NULL;
1054         struct md_op_data *op_data;
1055         int rc;
1056         ENTRY;
1057         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1058                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1059
1060         /*
1061          * XXX: unlink bind mountpoint maybe call to here,
1062          * just check it as vfs_unlink does.
1063          */
1064         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1065                 RETURN(-EBUSY);
1066
1067         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1068                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1069         if (IS_ERR(op_data))
1070                 RETURN(PTR_ERR(op_data));
1071
1072         ll_get_child_fid(dir, name, &op_data->op_fid3);
1073         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1074         ll_finish_md_op_data(op_data);
1075         if (rc)
1076                 GOTO(out, rc);
1077
1078         ll_update_times(request, dir);
1079         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1080
1081         rc = ll_objects_destroy(request, dir);
1082  out:
1083         ptlrpc_req_finished(request);
1084         RETURN(rc);
1085 }
1086
1087 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1088                              struct dentry *src_dchild, struct qstr *src_name,
1089                              struct inode *tgt, struct dentry *tgt_dparent,
1090                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1091 {
1092         struct ptlrpc_request *request = NULL;
1093         struct ll_sb_info *sbi = ll_i2sbi(src);
1094         struct md_op_data *op_data;
1095         int err;
1096         ENTRY;
1097         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1098                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1099                src->i_ino, src->i_generation, src, tgt_name->len,
1100                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1101
1102         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1103             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1104                 RETURN(-EBUSY);
1105
1106         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1107                                      LUSTRE_OPC_ANY, NULL);
1108         if (IS_ERR(op_data))
1109                 RETURN(PTR_ERR(op_data));
1110
1111         ll_get_child_fid(src, src_name, &op_data->op_fid3);
1112         ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1113         err = md_rename(sbi->ll_md_exp, op_data,
1114                         src_name->name, src_name->len,
1115                         tgt_name->name, tgt_name->len, &request);
1116         ll_finish_md_op_data(op_data);
1117         if (!err) {
1118                 ll_update_times(request, src);
1119                 ll_update_times(request, tgt);
1120                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1121                 err = ll_objects_destroy(request, src);
1122         }
1123
1124         ptlrpc_req_finished(request);
1125
1126         RETURN(err);
1127 }
1128
1129 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1130                     ll_dev_t rdev)
1131 {
1132         return ll_mknod_generic(dir, &dchild->d_name, mode,
1133                                 old_encode_dev(rdev), dchild);
1134 }
1135
1136 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1137 {
1138         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1139 }
1140 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1141 {
1142         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1143 }
1144 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1145 {
1146         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1147 }
1148 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1149                       const char *oldname)
1150 {
1151         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1152 }
1153 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1154                    struct dentry *new_dentry)
1155 {
1156         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1157                                new_dentry);
1158 }
1159 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1160                      struct inode *new_dir, struct dentry *new_dentry)
1161 {
1162         int err;
1163         err = ll_rename_generic(old_dir, NULL,
1164                                  old_dentry, &old_dentry->d_name,
1165                                  new_dir, NULL, new_dentry,
1166                                  &new_dentry->d_name);
1167         if (!err) {
1168 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1169                 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1170 #endif
1171                         d_move(old_dentry, new_dentry);
1172         }
1173         return err;
1174 }
1175
1176 struct inode_operations ll_dir_inode_operations = {
1177         .mknod              = ll_mknod,
1178         .lookup             = ll_lookup_nd,
1179         .create             = ll_create_nd,
1180         /* We need all these non-raw things for NFSD, to not patch it. */
1181         .unlink             = ll_unlink,
1182         .mkdir              = ll_mkdir,
1183         .rmdir              = ll_rmdir,
1184         .symlink            = ll_symlink,
1185         .link               = ll_link,
1186         .rename             = ll_rename,
1187         .setattr            = ll_setattr,
1188         .getattr            = ll_getattr,
1189         .permission         = ll_inode_permission,
1190         .setxattr           = ll_setxattr,
1191         .getxattr           = ll_getxattr,
1192         .listxattr          = ll_listxattr,
1193         .removexattr        = ll_removexattr,
1194 };
1195
1196 struct inode_operations ll_special_inode_operations = {
1197         .setattr        = ll_setattr,
1198         .getattr        = ll_getattr,
1199         .permission     = ll_inode_permission,
1200         .setxattr       = ll_setxattr,
1201         .getxattr       = ll_getxattr,
1202         .listxattr      = ll_listxattr,
1203         .removexattr    = ll_removexattr,
1204 };