Whamcloud - gitweb
LU-169 ldlm: add support for layout lock
[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
220                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
221                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
222                             MDS_INODELOCK_LAYOUT))
223                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
224
225                 if (bits & MDS_INODELOCK_OPEN)
226                         ll_have_md_lock(inode, &bits, mode);
227
228                 fid = ll_inode2fid(inode);
229                 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
230                     lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
231                     lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
232                         LDLM_ERROR(lock, "data mismatch with object "
233                                    DFID" (%p)", PFID(fid), inode);
234                 }
235
236                 if (bits & MDS_INODELOCK_OPEN) {
237                         int flags = 0;
238                         switch (lock->l_req_mode) {
239                         case LCK_CW:
240                                 flags = FMODE_WRITE;
241                                 break;
242                         case LCK_PR:
243                                 flags = FMODE_EXEC;
244                                 break;
245                         case LCK_CR:
246                                 flags = FMODE_READ;
247                                 break;
248                         default:
249                                 CERROR("Unexpected lock mode for OPEN lock "
250                                        "%d, inode %ld\n", lock->l_req_mode,
251                                        inode->i_ino);
252                         }
253                         ll_md_real_close(inode, flags);
254                 }
255
256                 lli = ll_i2info(inode);
257                 if (bits & MDS_INODELOCK_LAYOUT) {
258                         struct cl_object_conf conf = { .coc_inode = inode,
259                                                        .coc_invalidate = true };
260                         rc = ll_layout_conf(inode, &conf);
261                         if (rc)
262                                 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
263                 }
264
265                 if (bits & MDS_INODELOCK_UPDATE)
266                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
267
268                 if (S_ISDIR(inode->i_mode) &&
269                      (bits & MDS_INODELOCK_UPDATE)) {
270                         CDEBUG(D_INODE, "invalidating inode %lu\n",
271                                inode->i_ino);
272                         truncate_inode_pages(inode->i_mapping, 0);
273                         ll_invalidate_negative_children(inode);
274                 }
275
276                 if (inode->i_sb->s_root &&
277                     inode != inode->i_sb->s_root->d_inode &&
278                     (bits & MDS_INODELOCK_LOOKUP))
279                         ll_invalidate_aliases(inode);
280                 iput(inode);
281                 break;
282         }
283         default:
284                 LBUG();
285         }
286
287         RETURN(0);
288 }
289
290 __u32 ll_i2suppgid(struct inode *i)
291 {
292         if (cfs_curproc_is_in_groups(i->i_gid))
293                 return (__u32)i->i_gid;
294         else
295                 return (__u32)(-1);
296 }
297
298 /* Pack the required supplementary groups into the supplied groups array.
299  * If we don't need to use the groups from the target inode(s) then we
300  * instead pack one or more groups from the user's supplementary group
301  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
302 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
303 {
304 #if 0
305         int i;
306 #endif
307
308         LASSERT(i1 != NULL);
309         LASSERT(suppgids != NULL);
310
311         suppgids[0] = ll_i2suppgid(i1);
312
313         if (i2)
314                 suppgids[1] = ll_i2suppgid(i2);
315                 else
316                         suppgids[1] = -1;
317
318 #if 0
319         for (i = 0; i < current_ngroups; i++) {
320                 if (suppgids[0] == -1) {
321                         if (current_groups[i] != suppgids[1])
322                                 suppgids[0] = current_groups[i];
323                         continue;
324                 }
325                 if (suppgids[1] == -1) {
326                         if (current_groups[i] != suppgids[0])
327                                 suppgids[1] = current_groups[i];
328                         continue;
329                 }
330                 break;
331         }
332 #endif
333 }
334
335 /*
336  * try to reuse three types of dentry:
337  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
338  *    by concurrent .revalidate).
339  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
340  *    be cleared by others calling d_lustre_revalidate).
341  * 3. DISCONNECTED alias.
342  */
343 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
344 {
345         struct dentry *alias, *discon_alias, *invalid_alias;
346
347         if (list_empty(&inode->i_dentry))
348                 return NULL;
349
350         discon_alias = invalid_alias = NULL;
351
352         ll_lock_dcache(inode);
353         list_for_each_entry(alias, &inode->i_dentry, d_alias) {
354                 LASSERT(alias != dentry);
355
356                 spin_lock(&alias->d_lock);
357                 if (alias->d_flags & DCACHE_DISCONNECTED)
358                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
359                         discon_alias = alias;
360                 else if (alias->d_parent == dentry->d_parent             &&
361                          alias->d_name.hash == dentry->d_name.hash       &&
362                          alias->d_name.len == dentry->d_name.len         &&
363                          memcmp(alias->d_name.name, dentry->d_name.name,
364                                 dentry->d_name.len) == 0)
365                         invalid_alias = alias;
366                 spin_unlock(&alias->d_lock);
367
368                 if (invalid_alias)
369                         break;
370         }
371         alias = invalid_alias ?: discon_alias ?: NULL;
372         if (alias) {
373                 spin_lock(&alias->d_lock);
374                 dget_dlock(alias);
375                 spin_unlock(&alias->d_lock);
376         }
377         ll_unlock_dcache(inode);
378
379         return alias;
380 }
381
382 /*
383  * Similar to d_splice_alias(), but lustre treats DCACHE_LUSTRE_INVALID alias
384  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
385  */
386 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
387 {
388         struct dentry *new;
389
390         if (inode) {
391                 new = ll_find_alias(inode, de);
392                 if (new) {
393                         ll_dops_init(new, 1, 1);
394                         d_move(new, de);
395                         iput(inode);
396                         CDEBUG(D_DENTRY,
397                                "Reuse dentry %p inode %p refc %d flags %#x\n",
398                               new, new->d_inode, d_refcount(new), new->d_flags);
399                         return new;
400                 }
401         }
402         __d_lustre_invalidate(de);
403         ll_dops_init(de, 1, 1);
404         d_add(de, inode);
405         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
406                de, de->d_inode, d_refcount(de), de->d_flags);
407         return de;
408 }
409
410 int ll_lookup_it_finish(struct ptlrpc_request *request,
411                         struct lookup_intent *it, void *data)
412 {
413         struct it_cb_data *icbd = data;
414         struct dentry **de = icbd->icbd_childp;
415         struct inode *parent = icbd->icbd_parent;
416         struct inode *inode = NULL;
417         __u64 bits = 0;
418         int rc;
419         ENTRY;
420
421         /* NB 1 request reference will be taken away by ll_intent_lock()
422          * when I return */
423         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
424                 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
425                 if (rc)
426                         RETURN(rc);
427
428                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
429
430                 /* We used to query real size from OSTs here, but actually
431                    this is not needed. For stat() calls size would be updated
432                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
433                    2.4 and
434                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
435                    Everybody else who needs correct file size would call
436                    ll_glimpse_size or some equivalent themselves anyway.
437                    Also see bug 7198. */
438         }
439
440         *de = ll_splice_alias(inode, *de);
441
442         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
443                 /* we have lookup look - unhide dentry */
444                 if (bits & MDS_INODELOCK_LOOKUP)
445                         d_lustre_revalidate(*de);
446         } else {
447                 /* Check that parent has UPDATE lock. */
448                 struct lookup_intent parent_it = {
449                                         .it_op = IT_GETATTR,
450                                         .d.lustre.it_lock_handle = 0 };
451
452                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
453                                        &ll_i2info(parent)->lli_fid, NULL)) {
454                         d_lustre_revalidate(*de);
455                         ll_intent_release(&parent_it);
456                 }
457         }
458
459         RETURN(0);
460 }
461
462 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
463                                    struct lookup_intent *it, int lookup_flags)
464 {
465         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
466         struct dentry *save = dentry, *retval;
467         struct ptlrpc_request *req = NULL;
468         struct md_op_data *op_data;
469         struct it_cb_data icbd;
470         __u32 opc;
471         int rc;
472         ENTRY;
473
474         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
475                 RETURN(ERR_PTR(-ENAMETOOLONG));
476
477         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
478                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
479                parent->i_generation, parent, LL_IT2STR(it));
480
481         if (d_mountpoint(dentry))
482                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
483
484         ll_frob_intent(&it, &lookup_it);
485
486         /* As do_lookup is called before follow_mount, root dentry may be left
487          * not valid, revalidate it here. */
488         if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
489             (it->it_op & (IT_OPEN | IT_CREAT))) {
490                 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
491                                             MDS_INODELOCK_LOOKUP);
492                 if (rc)
493                         RETURN(ERR_PTR(rc));
494         }
495
496         if (it->it_op == IT_GETATTR) {
497                 rc = ll_statahead_enter(parent, &dentry, 0);
498                 if (rc == 1) {
499                         if (dentry == save)
500                                 GOTO(out, retval = NULL);
501                         GOTO(out, retval = dentry);
502                 }
503         }
504
505         icbd.icbd_childp = &dentry;
506         icbd.icbd_parent = parent;
507
508         if (it->it_op & IT_CREAT ||
509             (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
510                 opc = LUSTRE_OPC_CREATE;
511         else
512                 opc = LUSTRE_OPC_ANY;
513
514         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
515                                      dentry->d_name.len, lookup_flags, opc,
516                                      NULL);
517         if (IS_ERR(op_data))
518                 RETURN((void *)op_data);
519
520         it->it_create_mode &= ~cfs_curproc_umask();
521
522         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
523                             lookup_flags, &req, ll_md_blocking_ast, 0);
524         ll_finish_md_op_data(op_data);
525         if (rc < 0)
526                 GOTO(out, retval = ERR_PTR(rc));
527
528         rc = ll_lookup_it_finish(req, it, &icbd);
529         if (rc != 0) {
530                 ll_intent_release(it);
531                 GOTO(out, retval = ERR_PTR(rc));
532         }
533
534         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
535             !S_ISREG(dentry->d_inode->i_mode) &&
536             !S_ISDIR(dentry->d_inode->i_mode)) {
537                 ll_release_openhandle(dentry, it);
538         }
539         ll_lookup_finish_locks(it, dentry);
540
541         if (dentry == save)
542                 GOTO(out, retval = NULL);
543         else
544                 GOTO(out, retval = dentry);
545  out:
546         if (req)
547                 ptlrpc_req_finished(req);
548         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
549                 ll_statahead_mark(parent, dentry);
550         return retval;
551 }
552
553 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
554                                         int lookup_flags)
555 {
556         struct lookup_intent *it;
557
558         OBD_ALLOC(it, sizeof(*it));
559         if (!it)
560                 return ERR_PTR(-ENOMEM);
561
562         if (lookup_flags & LOOKUP_OPEN) {
563                 it->it_op = IT_OPEN;
564                 if (lookup_flags & LOOKUP_CREATE)
565                         it->it_op |= IT_CREAT;
566                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
567                 it->it_flags = oit->flags;
568         } else {
569                 it->it_op = IT_GETATTR;
570         }
571
572         return it;
573 }
574
575 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
576                                    struct nameidata *nd)
577 {
578         struct dentry *de;
579         ENTRY;
580
581         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
582                 struct lookup_intent *it;
583
584 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17)
585                 /* Did we came here from failed revalidate just to propagate
586                  * its error? */
587                 if (nd->flags & LOOKUP_OPEN)
588                         if (IS_ERR(nd->intent.open.file))
589                                 RETURN((struct dentry *)nd->intent.open.file);
590 #endif
591                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
592                         it = ll_d2d(dentry)->lld_it;
593                         ll_d2d(dentry)->lld_it = NULL;
594                 } else {
595                         if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
596                                 /* We are sure this is new dentry, so we need to create
597                                    our private data and set the dentry ops */ 
598                                 ll_dops_init(dentry, 1, 1);
599                                 __d_lustre_invalidate(dentry);
600                                 d_add(dentry, NULL);
601                                 RETURN(NULL);
602                         }
603                         it = ll_convert_intent(&nd->intent.open, nd->flags);
604                         if (IS_ERR(it))
605                                 RETURN((struct dentry *)it);
606                 }
607
608                 de = ll_lookup_it(parent, dentry, it, nd->flags);
609                 if (de)
610                         dentry = de;
611                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
612                         if (dentry->d_inode &&
613                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
614                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
615                                         // We cannot call open here as it would
616                                         // deadlock.
617                                         ptlrpc_req_finished(
618                                                        (struct ptlrpc_request *)
619                                                           it->d.lustre.it_data);
620                                 } else {
621                                         struct file *filp;
622
623                                         nd->intent.open.file->private_data = it;
624                                         filp = lookup_instantiate_filp(nd,
625                                                                        dentry,
626                                                                        NULL);
627                                         if (IS_ERR(filp)) {
628                                                 if (de)
629                                                         dput(de);
630                                                 de = (struct dentry *)filp;
631                                         }
632                                 }
633                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
634                                 // XXX This can only reliably work on assumption
635                                 // that there are NO hashed negative dentries.
636                                 ll_d2d(dentry)->lld_it = it;
637                                 it = NULL; /* Will be freed in ll_create_nd */
638                                 /* We absolutely depend on ll_create_nd to be
639                                  * called to not leak this intent and possible
640                                  * data attached to it */
641                         }
642                 }
643
644                 if (it) {
645                         ll_intent_release(it);
646                         OBD_FREE(it, sizeof(*it));
647                 }
648         } else {
649                 de = ll_lookup_it(parent, dentry, NULL, 0);
650         }
651
652         RETURN(de);
653 }
654
655 /* We depend on "mode" being set with the proper file type/umask by now */
656 static struct inode *ll_create_node(struct inode *dir, const char *name,
657                                     int namelen, const void *data, int datalen,
658                                     int mode, __u64 extra,
659                                     struct lookup_intent *it)
660 {
661         struct inode *inode = NULL;
662         struct ptlrpc_request *request = NULL;
663         struct ll_sb_info *sbi = ll_i2sbi(dir);
664         int rc;
665         ENTRY;
666
667         LASSERT(it && it->d.lustre.it_disposition);
668
669         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
670         request = it->d.lustre.it_data;
671         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
672         rc = ll_prep_inode(&inode, request, dir->i_sb);
673         if (rc)
674                 GOTO(out, inode = ERR_PTR(rc));
675
676         LASSERT(list_empty(&inode->i_dentry));
677
678         /* We asked for a lock on the directory, but were granted a
679          * lock on the inode.  Since we finally have an inode pointer,
680          * stuff it in the lock. */
681         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
682                inode, inode->i_ino, inode->i_generation);
683         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
684         EXIT;
685  out:
686         ptlrpc_req_finished(request);
687         return inode;
688 }
689
690 /*
691  * By the time this is called, we already have created the directory cache
692  * entry for the new file, but it is so far negative - it has no inode.
693  *
694  * We defer creating the OBD object(s) until open, to keep the intent and
695  * non-intent code paths similar, and also because we do not have the MDS
696  * inode number before calling ll_create_node() (which is needed for LOV),
697  * so we would need to do yet another RPC to the MDS to store the LOV EA
698  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
699  * lmm_size in datalen (the MDS still has code which will handle that).
700  *
701  * If the create succeeds, we fill in the inode information
702  * with d_instantiate().
703  */
704 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
705                         struct lookup_intent *it)
706 {
707         struct inode *inode;
708         int rc = 0;
709         ENTRY;
710
711         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
712                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
713                dir->i_generation, dir, LL_IT2STR(it));
714
715         rc = it_open_error(DISP_OPEN_CREATE, it);
716         if (rc)
717                 RETURN(rc);
718
719         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
720                                NULL, 0, mode, 0, it);
721         if (IS_ERR(inode))
722                 RETURN(PTR_ERR(inode));
723
724         d_instantiate(dentry, inode);
725         RETURN(0);
726 }
727
728 static void ll_update_times(struct ptlrpc_request *request,
729                             struct inode *inode)
730 {
731         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
732                                                        &RMF_MDT_BODY);
733
734         LASSERT(body);
735         if (body->valid & OBD_MD_FLMTIME &&
736             body->mtime > LTIME_S(inode->i_mtime)) {
737                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
738                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
739                 LTIME_S(inode->i_mtime) = body->mtime;
740         }
741         if (body->valid & OBD_MD_FLCTIME &&
742             body->ctime > LTIME_S(inode->i_ctime))
743                 LTIME_S(inode->i_ctime) = body->ctime;
744 }
745
746 static int ll_new_node(struct inode *dir, struct qstr *name,
747                        const char *tgt, int mode, int rdev,
748                        struct dentry *dchild, __u32 opc)
749 {
750         struct ptlrpc_request *request = NULL;
751         struct md_op_data *op_data;
752         struct inode *inode = NULL;
753         struct ll_sb_info *sbi = ll_i2sbi(dir);
754         int tgt_len = 0;
755         int err;
756
757         ENTRY;
758         if (unlikely(tgt != NULL))
759                 tgt_len = strlen(tgt) + 1;
760
761         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
762                                      name->len, 0, opc, NULL);
763         if (IS_ERR(op_data))
764                 GOTO(err_exit, err = PTR_ERR(op_data));
765
766         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
767                         cfs_curproc_fsuid(), cfs_curproc_fsgid(),
768                         cfs_curproc_cap_pack(), rdev, &request);
769         ll_finish_md_op_data(op_data);
770         if (err)
771                 GOTO(err_exit, err);
772
773         ll_update_times(request, dir);
774
775         if (dchild) {
776                 err = ll_prep_inode(&inode, request, dchild->d_sb);
777                 if (err)
778                      GOTO(err_exit, err);
779
780                 d_instantiate(dchild, inode);
781         }
782         EXIT;
783 err_exit:
784         ptlrpc_req_finished(request);
785
786         return err;
787 }
788
789 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
790                             unsigned rdev, struct dentry *dchild)
791 {
792         int err;
793         ENTRY;
794
795         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
796                name->len, name->name, dir->i_ino, dir->i_generation, dir,
797                mode, rdev);
798
799         mode &= ~cfs_curproc_umask();
800
801         switch (mode & S_IFMT) {
802         case 0:
803                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
804         case S_IFREG:
805         case S_IFCHR:
806         case S_IFBLK:
807         case S_IFIFO:
808         case S_IFSOCK:
809                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
810                                   LUSTRE_OPC_MKNOD);
811                 break;
812         case S_IFDIR:
813                 err = -EPERM;
814                 break;
815         default:
816                 err = -EINVAL;
817         }
818
819         if (!err)
820                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
821
822         RETURN(err);
823 }
824
825 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
826                         int mode, struct nameidata *nd)
827 {
828         struct lookup_intent *it = ll_d2d(dentry)->lld_it;
829         int rc;
830
831         if (!it)
832                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
833
834         ll_d2d(dentry)->lld_it = NULL;
835
836         /* Was there an error? Propagate it! */
837         if (it->d.lustre.it_status) {
838                 rc = it->d.lustre.it_status;
839                 goto out;
840         }
841
842         rc = ll_create_it(dir, dentry, mode, it);
843         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
844                 struct file *filp;
845
846                 nd->intent.open.file->private_data = it;
847                 filp = lookup_instantiate_filp(nd, dentry, NULL);
848                 if (IS_ERR(filp))
849                         rc = PTR_ERR(filp);
850         }
851
852 out:
853         ll_intent_release(it);
854         OBD_FREE(it, sizeof(*it));
855
856         if (!rc)
857                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
858
859         return rc;
860 }
861
862 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
863                               const char *tgt, struct dentry *dchild)
864 {
865         int err;
866         ENTRY;
867
868         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
869                name->len, name->name, dir->i_ino, dir->i_generation,
870                dir, 3000, tgt);
871
872         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
873                           0, dchild, LUSTRE_OPC_SYMLINK);
874
875         if (!err)
876                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
877
878         RETURN(err);
879 }
880
881 static int ll_link_generic(struct inode *src,  struct inode *dir,
882                            struct qstr *name, struct dentry *dchild)
883 {
884         struct ll_sb_info *sbi = ll_i2sbi(dir);
885         struct ptlrpc_request *request = NULL;
886         struct md_op_data *op_data;
887         int err;
888
889         ENTRY;
890         CDEBUG(D_VFSTRACE,
891                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
892                src->i_ino, src->i_generation, src, dir->i_ino,
893                dir->i_generation, dir, name->len, name->name);
894
895         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
896                                      0, LUSTRE_OPC_ANY, NULL);
897         if (IS_ERR(op_data))
898                 RETURN(PTR_ERR(op_data));
899
900         err = md_link(sbi->ll_md_exp, op_data, &request);
901         ll_finish_md_op_data(op_data);
902         if (err)
903                 GOTO(out, err);
904
905         ll_update_times(request, dir);
906         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
907         EXIT;
908 out:
909         ptlrpc_req_finished(request);
910         RETURN(err);
911 }
912
913 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
914                             int mode, struct dentry *dchild)
915
916 {
917         int err;
918         ENTRY;
919
920         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
921                name->len, name->name, dir->i_ino, dir->i_generation, dir);
922
923         mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
924         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
925
926         if (!err)
927                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
928
929         RETURN(err);
930 }
931
932 /* Try to find the child dentry by its name.
933    If found, put the result fid into @fid. */
934 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
935                              struct lu_fid *fid)
936 {
937         struct dentry *parent, *child;
938
939         parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
940         child = d_lookup(parent, name);
941         if (child) {
942                 if (child->d_inode)
943                         *fid = *ll_inode2fid(child->d_inode);
944                 dput(child);
945         }
946 }
947
948 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
949                             struct dentry *dchild, struct qstr *name)
950 {
951         struct ptlrpc_request *request = NULL;
952         struct md_op_data *op_data;
953         int rc;
954         ENTRY;
955
956         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
957                name->len, name->name, dir->i_ino, dir->i_generation, dir);
958
959         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
960                 RETURN(-EBUSY);
961
962         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
963                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
964         if (IS_ERR(op_data))
965                 RETURN(PTR_ERR(op_data));
966
967         ll_get_child_fid(dir, name, &op_data->op_fid3);
968         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
969         ll_finish_md_op_data(op_data);
970         if (rc == 0) {
971                 ll_update_times(request, dir);
972                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
973         }
974
975         ptlrpc_req_finished(request);
976         RETURN(rc);
977 }
978
979 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
980 {
981         struct mdt_body *body;
982         struct lov_mds_md *eadata;
983         struct lov_stripe_md *lsm = NULL;
984         struct obd_trans_info oti = { 0 };
985         struct obdo *oa;
986         struct obd_capa *oc = NULL;
987         int rc;
988         ENTRY;
989
990         /* req is swabbed so this is safe */
991         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
992         if (!(body->valid & OBD_MD_FLEASIZE))
993                 RETURN(0);
994
995         if (body->eadatasize == 0) {
996                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
997                 GOTO(out, rc = -EPROTO);
998         }
999
1000         /* The MDS sent back the EA because we unlinked the last reference
1001          * to this file. Use this EA to unlink the objects on the OST.
1002          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1003          * check it is complete and sensible. */
1004         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1005                                               body->eadatasize);
1006         LASSERT(eadata != NULL);
1007
1008         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1009         if (rc < 0) {
1010                 CERROR("obd_unpackmd: %d\n", rc);
1011                 GOTO(out, rc);
1012         }
1013         LASSERT(rc >= sizeof(*lsm));
1014
1015         OBDO_ALLOC(oa);
1016         if (oa == NULL)
1017                 GOTO(out_free_memmd, rc = -ENOMEM);
1018
1019         oa->o_id = lsm->lsm_object_id;
1020         oa->o_seq = lsm->lsm_object_seq;
1021         oa->o_mode = body->mode & S_IFMT;
1022         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1023
1024         if (body->valid & OBD_MD_FLCOOKIE) {
1025                 oa->o_valid |= OBD_MD_FLCOOKIE;
1026                 oti.oti_logcookies =
1027                         req_capsule_server_sized_get(&request->rq_pill,
1028                                                      &RMF_LOGCOOKIES,
1029                                                    sizeof(struct llog_cookie) *
1030                                                      lsm->lsm_stripe_count);
1031                 if (oti.oti_logcookies == NULL) {
1032                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1033                         body->valid &= ~OBD_MD_FLCOOKIE;
1034                 }
1035         }
1036
1037         if (body->valid & OBD_MD_FLOSSCAPA) {
1038                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1039                 if (rc)
1040                         GOTO(out_free_memmd, rc);
1041         }
1042
1043         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1044                          ll_i2mdexp(dir), oc);
1045         capa_put(oc);
1046         OBDO_FREE(oa);
1047         if (rc)
1048                 CERROR("obd destroy objid "LPX64" error %d\n",
1049                        lsm->lsm_object_id, rc);
1050  out_free_memmd:
1051         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1052  out:
1053         return rc;
1054 }
1055
1056 /* ll_unlink_generic() doesn't update the inode with the new link count.
1057  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1058  * is any lock existing. They will recycle dentries and inodes based upon locks
1059  * too. b=20433 */
1060 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1061                              struct dentry *dchild, struct qstr *name)
1062 {
1063         struct ptlrpc_request *request = NULL;
1064         struct md_op_data *op_data;
1065         int rc;
1066         ENTRY;
1067         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1068                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1069
1070         /*
1071          * XXX: unlink bind mountpoint maybe call to here,
1072          * just check it as vfs_unlink does.
1073          */
1074         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1075                 RETURN(-EBUSY);
1076
1077         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1078                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1079         if (IS_ERR(op_data))
1080                 RETURN(PTR_ERR(op_data));
1081
1082         ll_get_child_fid(dir, name, &op_data->op_fid3);
1083         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1084         ll_finish_md_op_data(op_data);
1085         if (rc)
1086                 GOTO(out, rc);
1087
1088         ll_update_times(request, dir);
1089         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1090
1091         rc = ll_objects_destroy(request, dir);
1092  out:
1093         ptlrpc_req_finished(request);
1094         RETURN(rc);
1095 }
1096
1097 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1098                              struct dentry *src_dchild, struct qstr *src_name,
1099                              struct inode *tgt, struct dentry *tgt_dparent,
1100                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1101 {
1102         struct ptlrpc_request *request = NULL;
1103         struct ll_sb_info *sbi = ll_i2sbi(src);
1104         struct md_op_data *op_data;
1105         int err;
1106         ENTRY;
1107         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1108                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1109                src->i_ino, src->i_generation, src, tgt_name->len,
1110                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1111
1112         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1113             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1114                 RETURN(-EBUSY);
1115
1116         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1117                                      LUSTRE_OPC_ANY, NULL);
1118         if (IS_ERR(op_data))
1119                 RETURN(PTR_ERR(op_data));
1120
1121         ll_get_child_fid(src, src_name, &op_data->op_fid3);
1122         ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1123         err = md_rename(sbi->ll_md_exp, op_data,
1124                         src_name->name, src_name->len,
1125                         tgt_name->name, tgt_name->len, &request);
1126         ll_finish_md_op_data(op_data);
1127         if (!err) {
1128                 ll_update_times(request, src);
1129                 ll_update_times(request, tgt);
1130                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1131                 err = ll_objects_destroy(request, src);
1132         }
1133
1134         ptlrpc_req_finished(request);
1135
1136         RETURN(err);
1137 }
1138
1139 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1140                     ll_dev_t rdev)
1141 {
1142         return ll_mknod_generic(dir, &dchild->d_name, mode,
1143                                 old_encode_dev(rdev), dchild);
1144 }
1145
1146 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1147 {
1148         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1149 }
1150 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1151 {
1152         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1153 }
1154 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1155 {
1156         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1157 }
1158 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1159                       const char *oldname)
1160 {
1161         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1162 }
1163 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1164                    struct dentry *new_dentry)
1165 {
1166         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1167                                new_dentry);
1168 }
1169 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1170                      struct inode *new_dir, struct dentry *new_dentry)
1171 {
1172         int err;
1173         err = ll_rename_generic(old_dir, NULL,
1174                                  old_dentry, &old_dentry->d_name,
1175                                  new_dir, NULL, new_dentry,
1176                                  &new_dentry->d_name);
1177         if (!err) {
1178 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1179                 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1180 #endif
1181                         d_move(old_dentry, new_dentry);
1182         }
1183         return err;
1184 }
1185
1186 struct inode_operations ll_dir_inode_operations = {
1187         .mknod              = ll_mknod,
1188         .lookup             = ll_lookup_nd,
1189         .create             = ll_create_nd,
1190         /* We need all these non-raw things for NFSD, to not patch it. */
1191         .unlink             = ll_unlink,
1192         .mkdir              = ll_mkdir,
1193         .rmdir              = ll_rmdir,
1194         .symlink            = ll_symlink,
1195         .link               = ll_link,
1196         .rename             = ll_rename,
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 };
1205
1206 struct inode_operations ll_special_inode_operations = {
1207         .setattr        = ll_setattr,
1208         .getattr        = ll_getattr,
1209         .permission     = ll_inode_permission,
1210         .setxattr       = ll_setxattr,
1211         .getxattr       = ll_getxattr,
1212         .listxattr      = ll_listxattr,
1213         .removexattr    = ll_removexattr,
1214 };