Whamcloud - gitweb
6fa058c705472b570a5819027fdb0c26866313cb
[fs/lustre-release.git] / lustre / llite / namei.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  */
36
37 #include <linux/fs.h>
38 #include <linux/sched.h>
39 #include <linux/mm.h>
40 #include <linux/quotaops.h>
41 #include <linux/highmem.h>
42 #include <linux/pagemap.h>
43 #include <linux/security.h>
44
45 #define DEBUG_SUBSYSTEM S_LLITE
46
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <lustre_lite.h>
50 #include <lustre_dlm.h>
51 #include <lustre_ver.h>
52 #include "llite_internal.h"
53
54 /*
55  * Check if we have something mounted at the named dchild.
56  * In such a case there would always be dentry present.
57  */
58 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
59                            struct qstr *name)
60 {
61         int mounted = 0;
62
63         if (unlikely(dchild)) {
64                 mounted = d_mountpoint(dchild);
65         } else if (dparent) {
66                 dchild = d_lookup(dparent, name);
67                 if (dchild) {
68                         mounted = d_mountpoint(dchild);
69                         dput(dchild);
70                 }
71         }
72         return mounted;
73 }
74
75 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
76 {
77         ENTRY;
78
79         ldlm_lock_decref(lockh, mode);
80
81         RETURN(0);
82 }
83
84
85 /* called from iget5_locked->find_inode() under inode_lock spinlock */
86 static int ll_test_inode(struct inode *inode, void *opaque)
87 {
88         struct ll_inode_info *lli = ll_i2info(inode);
89         struct lustre_md     *md = opaque;
90
91         if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
92                 CERROR("MDS body missing FID\n");
93                 return 0;
94         }
95
96         if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
97                 return 0;
98
99         return 1;
100 }
101
102 static int ll_set_inode(struct inode *inode, void *opaque)
103 {
104         struct ll_inode_info *lli = ll_i2info(inode);
105         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
106
107         if (unlikely(!(body->valid & OBD_MD_FLID))) {
108                 CERROR("MDS body missing FID\n");
109                 return -EINVAL;
110         }
111
112         lli->lli_fid = body->fid1;
113         if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
114                 CERROR("Can not initialize inode "DFID" without object type: "
115                        "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
116                 return -EINVAL;
117         }
118
119         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
120         if (unlikely(inode->i_mode == 0)) {
121                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
122                 return -EINVAL;
123         }
124
125         ll_lli_init(lli);
126
127         return 0;
128 }
129
130
131 /*
132  * Get an inode by inode number (already instantiated by the intent lookup).
133  * Returns inode or NULL
134  */
135 struct inode *ll_iget(struct super_block *sb, ino_t hash,
136                       struct lustre_md *md)
137 {
138         struct inode         *inode;
139         ENTRY;
140
141         LASSERT(hash != 0);
142         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
143
144         if (inode) {
145                 if (inode->i_state & I_NEW) {
146                         int rc = 0;
147
148                         ll_read_inode2(inode, md);
149                         if (S_ISREG(inode->i_mode) &&
150                             ll_i2info(inode)->lli_clob == NULL)
151                                 rc = cl_file_inode_init(inode, md);
152                         if (rc != 0) {
153                                 md->lsm = NULL;
154                                 make_bad_inode(inode);
155                                 unlock_new_inode(inode);
156                                 iput(inode);
157                                 inode = ERR_PTR(rc);
158                         } else
159                                 unlock_new_inode(inode);
160                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
161                         ll_update_inode(inode, md);
162                 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
163                        inode, PFID(&md->body->fid1));
164         }
165         RETURN(inode);
166 }
167
168 static void ll_invalidate_negative_children(struct inode *dir)
169 {
170         struct dentry *dentry, *tmp_subdir;
171         struct ll_d_hlist_node *p;
172
173         ll_lock_dcache(dir);
174         ll_d_hlist_for_each_entry(dentry, p, &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_u.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         struct ll_d_hlist_node *p;
347
348         if (ll_d_hlist_empty(&inode->i_dentry))
349                 return NULL;
350
351         discon_alias = invalid_alias = NULL;
352
353         ll_lock_dcache(inode);
354         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
355                 LASSERT(alias != dentry);
356
357                 spin_lock(&alias->d_lock);
358                 if (alias->d_flags & DCACHE_DISCONNECTED)
359                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
360                         discon_alias = alias;
361                 else if (alias->d_parent == dentry->d_parent             &&
362                          alias->d_name.hash == dentry->d_name.hash       &&
363                          alias->d_name.len == dentry->d_name.len         &&
364                          memcmp(alias->d_name.name, dentry->d_name.name,
365                                 dentry->d_name.len) == 0)
366                         invalid_alias = alias;
367                 spin_unlock(&alias->d_lock);
368
369                 if (invalid_alias)
370                         break;
371         }
372         alias = invalid_alias ?: discon_alias ?: NULL;
373         if (alias) {
374                 spin_lock(&alias->d_lock);
375                 dget_dlock(alias);
376                 spin_unlock(&alias->d_lock);
377         }
378         ll_unlock_dcache(inode);
379
380         return alias;
381 }
382
383 /*
384  * Similar to d_splice_alias(), but lustre treats invalid alias
385  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
386  */
387 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
388 {
389         struct dentry *new;
390
391         if (inode) {
392                 new = ll_find_alias(inode, de);
393                 if (new) {
394                         ll_dops_init(new, 1, 1);
395                         d_move(new, de);
396                         iput(inode);
397                         CDEBUG(D_DENTRY,
398                                "Reuse dentry %p inode %p refc %d flags %#x\n",
399                               new, new->d_inode, d_refcount(new), new->d_flags);
400                         return new;
401                 }
402         }
403         ll_dops_init(de, 1, 1);
404         __d_lustre_invalidate(de);
405         d_add(de, inode);
406         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
407                de, de->d_inode, d_refcount(de), de->d_flags);
408         return de;
409 }
410
411 int ll_lookup_it_finish(struct ptlrpc_request *request,
412                         struct lookup_intent *it, void *data)
413 {
414         struct it_cb_data *icbd = data;
415         struct dentry **de = icbd->icbd_childp;
416         struct inode *parent = icbd->icbd_parent;
417         struct inode *inode = NULL;
418         __u64 bits = 0;
419         int rc;
420         ENTRY;
421
422         /* NB 1 request reference will be taken away by ll_intent_lock()
423          * when I return */
424         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
425                 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
426                 if (rc)
427                         RETURN(rc);
428
429                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
430
431                 /* We used to query real size from OSTs here, but actually
432                    this is not needed. For stat() calls size would be updated
433                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
434                    2.4 and
435                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
436                    Everybody else who needs correct file size would call
437                    ll_glimpse_size or some equivalent themselves anyway.
438                    Also see bug 7198. */
439         }
440
441         *de = ll_splice_alias(inode, *de);
442
443         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
444                 /* we have lookup look - unhide dentry */
445                 if (bits & MDS_INODELOCK_LOOKUP)
446                         d_lustre_revalidate(*de);
447         } else {
448                 /* Check that parent has UPDATE lock. */
449                 struct lookup_intent parent_it = {
450                                         .it_op = IT_GETATTR,
451                                         .d.lustre.it_lock_handle = 0 };
452
453                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
454                                        &ll_i2info(parent)->lli_fid, NULL)) {
455                         d_lustre_revalidate(*de);
456                         ll_intent_release(&parent_it);
457                 }
458         }
459
460         RETURN(0);
461 }
462
463 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
464                                    struct lookup_intent *it, int lookup_flags)
465 {
466         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
467         struct dentry *save = dentry, *retval;
468         struct ptlrpc_request *req = NULL;
469         struct md_op_data *op_data;
470         struct it_cb_data icbd;
471         __u32 opc;
472         int rc;
473         ENTRY;
474
475         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
476                 RETURN(ERR_PTR(-ENAMETOOLONG));
477
478         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
479                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
480                parent->i_generation, parent, LL_IT2STR(it));
481
482         if (d_mountpoint(dentry))
483                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
484
485         ll_frob_intent(&it, &lookup_it);
486
487         /* As do_lookup is called before follow_mount, root dentry may be left
488          * not valid, revalidate it here. */
489         if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
490             (it->it_op & (IT_OPEN | IT_CREAT))) {
491                 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
492                                             MDS_INODELOCK_LOOKUP);
493                 if (rc)
494                         RETURN(ERR_PTR(rc));
495         }
496
497         if (it->it_op == IT_GETATTR) {
498                 rc = ll_statahead_enter(parent, &dentry, 0);
499                 if (rc == 1) {
500                         if (dentry == save)
501                                 GOTO(out, retval = NULL);
502                         GOTO(out, retval = dentry);
503                 }
504         }
505
506         icbd.icbd_childp = &dentry;
507         icbd.icbd_parent = parent;
508
509         if (it->it_op & IT_CREAT ||
510             (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
511                 opc = LUSTRE_OPC_CREATE;
512         else
513                 opc = LUSTRE_OPC_ANY;
514
515         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
516                                      dentry->d_name.len, lookup_flags, opc,
517                                      NULL);
518         if (IS_ERR(op_data))
519                 RETURN((void *)op_data);
520
521         /* enforce umask if acl disabled or MDS doesn't support umask */
522         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
523                 it->it_create_mode &= ~cfs_curproc_umask();
524
525         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
526                             lookup_flags, &req, ll_md_blocking_ast, 0);
527         ll_finish_md_op_data(op_data);
528         if (rc < 0)
529                 GOTO(out, retval = ERR_PTR(rc));
530
531         rc = ll_lookup_it_finish(req, it, &icbd);
532         if (rc != 0) {
533                 ll_intent_release(it);
534                 GOTO(out, retval = ERR_PTR(rc));
535         }
536
537         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
538             !S_ISREG(dentry->d_inode->i_mode) &&
539             !S_ISDIR(dentry->d_inode->i_mode)) {
540                 ll_release_openhandle(dentry, it);
541         }
542         ll_lookup_finish_locks(it, dentry);
543
544         if (dentry == save)
545                 GOTO(out, retval = NULL);
546         else
547                 GOTO(out, retval = dentry);
548  out:
549         if (req)
550                 ptlrpc_req_finished(req);
551         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
552                 ll_statahead_mark(parent, dentry);
553         return retval;
554 }
555
556 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
557                                         int lookup_flags)
558 {
559         struct lookup_intent *it;
560
561         OBD_ALLOC(it, sizeof(*it));
562         if (!it)
563                 return ERR_PTR(-ENOMEM);
564
565         if (lookup_flags & LOOKUP_OPEN) {
566                 it->it_op = IT_OPEN;
567                 if (lookup_flags & LOOKUP_CREATE)
568                         it->it_op |= IT_CREAT;
569                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
570                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
571         } else {
572                 it->it_op = IT_GETATTR;
573         }
574
575         return it;
576 }
577
578 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
579                                    struct nameidata *nd)
580 {
581         struct dentry *de;
582         ENTRY;
583
584         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
585                 struct lookup_intent *it;
586
587                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
588                         it = ll_d2d(dentry)->lld_it;
589                         ll_d2d(dentry)->lld_it = NULL;
590                 } else {
591                         if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
592                                 /* We are sure this is new dentry, so we need to create
593                                    our private data and set the dentry ops */ 
594                                 ll_dops_init(dentry, 1, 1);
595                                 __d_lustre_invalidate(dentry);
596                                 d_add(dentry, NULL);
597                                 RETURN(NULL);
598                         }
599                         it = ll_convert_intent(&nd->intent.open, nd->flags);
600                         if (IS_ERR(it))
601                                 RETURN((struct dentry *)it);
602                 }
603
604                 de = ll_lookup_it(parent, dentry, it, nd->flags);
605                 if (de)
606                         dentry = de;
607                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
608                         if (dentry->d_inode &&
609                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
610                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
611                                         // We cannot call open here as it would
612                                         // deadlock.
613                                         ptlrpc_req_finished(
614                                                        (struct ptlrpc_request *)
615                                                           it->d.lustre.it_data);
616                                 } else {
617                                         struct file *filp;
618
619                                         nd->intent.open.file->private_data = it;
620                                         filp = lookup_instantiate_filp(nd,
621                                                                        dentry,
622                                                                        NULL);
623                                         if (IS_ERR(filp)) {
624                                                 if (de)
625                                                         dput(de);
626                                                 de = (struct dentry *)filp;
627                                         }
628                                 }
629                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
630                                 // XXX This can only reliably work on assumption
631                                 // that there are NO hashed negative dentries.
632                                 ll_d2d(dentry)->lld_it = it;
633                                 it = NULL; /* Will be freed in ll_create_nd */
634                                 /* We absolutely depend on ll_create_nd to be
635                                  * called to not leak this intent and possible
636                                  * data attached to it */
637                         }
638                 }
639
640                 if (it) {
641                         ll_intent_release(it);
642                         OBD_FREE(it, sizeof(*it));
643                 }
644         } else {
645                 de = ll_lookup_it(parent, dentry, NULL, 0);
646         }
647
648         RETURN(de);
649 }
650
651 /* We depend on "mode" being set with the proper file type/umask by now */
652 static struct inode *ll_create_node(struct inode *dir, const char *name,
653                                     int namelen, const void *data, int datalen,
654                                     int mode, __u64 extra,
655                                     struct lookup_intent *it)
656 {
657         struct inode *inode = NULL;
658         struct ptlrpc_request *request = NULL;
659         struct ll_sb_info *sbi = ll_i2sbi(dir);
660         int rc;
661         ENTRY;
662
663         LASSERT(it && it->d.lustre.it_disposition);
664
665         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
666         request = it->d.lustre.it_data;
667         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
668         rc = ll_prep_inode(&inode, request, dir->i_sb);
669         if (rc)
670                 GOTO(out, inode = ERR_PTR(rc));
671
672         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
673
674         /* We asked for a lock on the directory, but were granted a
675          * lock on the inode.  Since we finally have an inode pointer,
676          * stuff it in the lock. */
677         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
678                inode, inode->i_ino, inode->i_generation);
679         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
680         EXIT;
681  out:
682         ptlrpc_req_finished(request);
683         return inode;
684 }
685
686 /*
687  * By the time this is called, we already have created the directory cache
688  * entry for the new file, but it is so far negative - it has no inode.
689  *
690  * We defer creating the OBD object(s) until open, to keep the intent and
691  * non-intent code paths similar, and also because we do not have the MDS
692  * inode number before calling ll_create_node() (which is needed for LOV),
693  * so we would need to do yet another RPC to the MDS to store the LOV EA
694  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
695  * lmm_size in datalen (the MDS still has code which will handle that).
696  *
697  * If the create succeeds, we fill in the inode information
698  * with d_instantiate().
699  */
700 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
701                         struct lookup_intent *it)
702 {
703         struct inode *inode;
704         int rc = 0;
705         ENTRY;
706
707         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
708                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
709                dir->i_generation, dir, LL_IT2STR(it));
710
711         rc = it_open_error(DISP_OPEN_CREATE, it);
712         if (rc)
713                 RETURN(rc);
714
715         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
716                                NULL, 0, mode, 0, it);
717         if (IS_ERR(inode))
718                 RETURN(PTR_ERR(inode));
719
720         d_instantiate(dentry, inode);
721         RETURN(0);
722 }
723
724 static void ll_update_times(struct ptlrpc_request *request,
725                             struct inode *inode)
726 {
727         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
728                                                        &RMF_MDT_BODY);
729
730         LASSERT(body);
731         if (body->valid & OBD_MD_FLMTIME &&
732             body->mtime > LTIME_S(inode->i_mtime)) {
733                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
734                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
735                 LTIME_S(inode->i_mtime) = body->mtime;
736         }
737         if (body->valid & OBD_MD_FLCTIME &&
738             body->ctime > LTIME_S(inode->i_ctime))
739                 LTIME_S(inode->i_ctime) = body->ctime;
740 }
741
742 static int ll_new_node(struct inode *dir, struct qstr *name,
743                        const char *tgt, int mode, int rdev,
744                        struct dentry *dchild, __u32 opc)
745 {
746         struct ptlrpc_request *request = NULL;
747         struct md_op_data *op_data;
748         struct inode *inode = NULL;
749         struct ll_sb_info *sbi = ll_i2sbi(dir);
750         int tgt_len = 0;
751         int err;
752
753         ENTRY;
754         if (unlikely(tgt != NULL))
755                 tgt_len = strlen(tgt) + 1;
756
757         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
758                                      name->len, 0, opc, NULL);
759         if (IS_ERR(op_data))
760                 GOTO(err_exit, err = PTR_ERR(op_data));
761
762         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
763                         cfs_curproc_fsuid(), cfs_curproc_fsgid(),
764                         cfs_curproc_cap_pack(), rdev, &request);
765         ll_finish_md_op_data(op_data);
766         if (err)
767                 GOTO(err_exit, err);
768
769         ll_update_times(request, dir);
770
771         if (dchild) {
772                 err = ll_prep_inode(&inode, request, dchild->d_sb);
773                 if (err)
774                      GOTO(err_exit, err);
775
776                 d_instantiate(dchild, inode);
777         }
778         EXIT;
779 err_exit:
780         ptlrpc_req_finished(request);
781
782         return err;
783 }
784
785 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
786                             unsigned rdev, struct dentry *dchild)
787 {
788         int err;
789         ENTRY;
790
791         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
792                name->len, name->name, dir->i_ino, dir->i_generation, dir,
793                mode, rdev);
794
795         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
796                 mode &= ~cfs_curproc_umask();
797
798         switch (mode & S_IFMT) {
799         case 0:
800                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
801         case S_IFREG:
802         case S_IFCHR:
803         case S_IFBLK:
804         case S_IFIFO:
805         case S_IFSOCK:
806                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
807                                   LUSTRE_OPC_MKNOD);
808                 break;
809         case S_IFDIR:
810                 err = -EPERM;
811                 break;
812         default:
813                 err = -EINVAL;
814         }
815
816         if (!err)
817                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
818
819         RETURN(err);
820 }
821
822 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
823                         ll_umode_t mode, struct nameidata *nd)
824 {
825         struct lookup_intent *it = ll_d2d(dentry)->lld_it;
826         int rc;
827
828         if (!it)
829                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
830
831         ll_d2d(dentry)->lld_it = NULL;
832
833         /* Was there an error? Propagate it! */
834         if (it->d.lustre.it_status) {
835                 rc = it->d.lustre.it_status;
836                 goto out;
837         }
838
839         rc = ll_create_it(dir, dentry, mode, it);
840         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
841                 struct file *filp;
842
843                 nd->intent.open.file->private_data = it;
844                 filp = lookup_instantiate_filp(nd, dentry, NULL);
845                 if (IS_ERR(filp))
846                         rc = PTR_ERR(filp);
847         }
848
849 out:
850         ll_intent_release(it);
851         OBD_FREE(it, sizeof(*it));
852
853         if (!rc)
854                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
855
856         return rc;
857 }
858
859 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
860                               const char *tgt, struct dentry *dchild)
861 {
862         int err;
863         ENTRY;
864
865         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
866                name->len, name->name, dir->i_ino, dir->i_generation,
867                dir, 3000, tgt);
868
869         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
870                           0, dchild, LUSTRE_OPC_SYMLINK);
871
872         if (!err)
873                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
874
875         RETURN(err);
876 }
877
878 static int ll_link_generic(struct inode *src,  struct inode *dir,
879                            struct qstr *name, struct dentry *dchild)
880 {
881         struct ll_sb_info *sbi = ll_i2sbi(dir);
882         struct ptlrpc_request *request = NULL;
883         struct md_op_data *op_data;
884         int err;
885
886         ENTRY;
887         CDEBUG(D_VFSTRACE,
888                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
889                src->i_ino, src->i_generation, src, dir->i_ino,
890                dir->i_generation, dir, name->len, name->name);
891
892         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
893                                      0, LUSTRE_OPC_ANY, NULL);
894         if (IS_ERR(op_data))
895                 RETURN(PTR_ERR(op_data));
896
897         err = md_link(sbi->ll_md_exp, op_data, &request);
898         ll_finish_md_op_data(op_data);
899         if (err)
900                 GOTO(out, err);
901
902         ll_update_times(request, dir);
903         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
904         EXIT;
905 out:
906         ptlrpc_req_finished(request);
907         RETURN(err);
908 }
909
910 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
911                             int mode, struct dentry *dchild)
912
913 {
914         int err;
915         ENTRY;
916
917         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
918                name->len, name->name, dir->i_ino, dir->i_generation, dir);
919
920         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
921                 mode &= ~cfs_curproc_umask();
922         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
923         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
924
925         if (!err)
926                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
927
928         RETURN(err);
929 }
930
931 /* Try to find the child dentry by its name.
932    If found, put the result fid into @fid. */
933 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
934                              struct lu_fid *fid)
935 {
936         struct dentry *parent, *child;
937
938         parent = ll_d_hlist_entry(dir->i_dentry, struct dentry, d_alias);
939         child = d_lookup(parent, name);
940         if (child) {
941                 if (child->d_inode)
942                         *fid = *ll_inode2fid(child->d_inode);
943                 dput(child);
944         }
945 }
946
947 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
948                             struct dentry *dchild, struct qstr *name)
949 {
950         struct ptlrpc_request *request = NULL;
951         struct md_op_data *op_data;
952         int rc;
953         ENTRY;
954
955         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
956                name->len, name->name, dir->i_ino, dir->i_generation, dir);
957
958         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
959                 RETURN(-EBUSY);
960
961         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
962                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
963         if (IS_ERR(op_data))
964                 RETURN(PTR_ERR(op_data));
965
966         ll_get_child_fid(dir, name, &op_data->op_fid3);
967         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
968         ll_finish_md_op_data(op_data);
969         if (rc == 0) {
970                 ll_update_times(request, dir);
971                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
972         }
973
974         ptlrpc_req_finished(request);
975         RETURN(rc);
976 }
977
978 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
979 {
980         struct mdt_body *body;
981         struct lov_mds_md *eadata;
982         struct lov_stripe_md *lsm = NULL;
983         struct obd_trans_info oti = { 0 };
984         struct obdo *oa;
985         struct obd_capa *oc = NULL;
986         int rc;
987         ENTRY;
988
989         /* req is swabbed so this is safe */
990         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
991         if (!(body->valid & OBD_MD_FLEASIZE))
992                 RETURN(0);
993
994         if (body->eadatasize == 0) {
995                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
996                 GOTO(out, rc = -EPROTO);
997         }
998
999         /* The MDS sent back the EA because we unlinked the last reference
1000          * to this file. Use this EA to unlink the objects on the OST.
1001          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1002          * check it is complete and sensible. */
1003         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1004                                               body->eadatasize);
1005         LASSERT(eadata != NULL);
1006
1007         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1008         if (rc < 0) {
1009                 CERROR("obd_unpackmd: %d\n", rc);
1010                 GOTO(out, rc);
1011         }
1012         LASSERT(rc >= sizeof(*lsm));
1013
1014         OBDO_ALLOC(oa);
1015         if (oa == NULL)
1016                 GOTO(out_free_memmd, rc = -ENOMEM);
1017
1018         oa->o_id = lsm->lsm_object_id;
1019         oa->o_seq = lsm->lsm_object_seq;
1020         oa->o_mode = body->mode & S_IFMT;
1021         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1022
1023         if (body->valid & OBD_MD_FLCOOKIE) {
1024                 oa->o_valid |= OBD_MD_FLCOOKIE;
1025                 oti.oti_logcookies =
1026                         req_capsule_server_sized_get(&request->rq_pill,
1027                                                      &RMF_LOGCOOKIES,
1028                                                    sizeof(struct llog_cookie) *
1029                                                      lsm->lsm_stripe_count);
1030                 if (oti.oti_logcookies == NULL) {
1031                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1032                         body->valid &= ~OBD_MD_FLCOOKIE;
1033                 }
1034         }
1035
1036         if (body->valid & OBD_MD_FLOSSCAPA) {
1037                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1038                 if (rc)
1039                         GOTO(out_free_memmd, rc);
1040         }
1041
1042         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1043                          ll_i2mdexp(dir), oc);
1044         capa_put(oc);
1045         OBDO_FREE(oa);
1046         if (rc)
1047                 CERROR("obd destroy objid "LPX64" error %d\n",
1048                        lsm->lsm_object_id, rc);
1049  out_free_memmd:
1050         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1051  out:
1052         return rc;
1053 }
1054
1055 /* ll_unlink_generic() doesn't update the inode with the new link count.
1056  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1057  * is any lock existing. They will recycle dentries and inodes based upon locks
1058  * too. b=20433 */
1059 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1060                              struct dentry *dchild, struct qstr *name)
1061 {
1062         struct ptlrpc_request *request = NULL;
1063         struct md_op_data *op_data;
1064         int rc;
1065         ENTRY;
1066         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1067                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1068
1069         /*
1070          * XXX: unlink bind mountpoint maybe call to here,
1071          * just check it as vfs_unlink does.
1072          */
1073         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1074                 RETURN(-EBUSY);
1075
1076         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1077                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1078         if (IS_ERR(op_data))
1079                 RETURN(PTR_ERR(op_data));
1080
1081         ll_get_child_fid(dir, name, &op_data->op_fid3);
1082         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1083         ll_finish_md_op_data(op_data);
1084         if (rc)
1085                 GOTO(out, rc);
1086
1087         ll_update_times(request, dir);
1088         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1089
1090         rc = ll_objects_destroy(request, dir);
1091  out:
1092         ptlrpc_req_finished(request);
1093         RETURN(rc);
1094 }
1095
1096 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1097                              struct dentry *src_dchild, struct qstr *src_name,
1098                              struct inode *tgt, struct dentry *tgt_dparent,
1099                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1100 {
1101         struct ptlrpc_request *request = NULL;
1102         struct ll_sb_info *sbi = ll_i2sbi(src);
1103         struct md_op_data *op_data;
1104         int err;
1105         ENTRY;
1106         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1107                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1108                src->i_ino, src->i_generation, src, tgt_name->len,
1109                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1110
1111         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1112             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1113                 RETURN(-EBUSY);
1114
1115         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1116                                      LUSTRE_OPC_ANY, NULL);
1117         if (IS_ERR(op_data))
1118                 RETURN(PTR_ERR(op_data));
1119
1120         ll_get_child_fid(src, src_name, &op_data->op_fid3);
1121         ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1122         err = md_rename(sbi->ll_md_exp, op_data,
1123                         src_name->name, src_name->len,
1124                         tgt_name->name, tgt_name->len, &request);
1125         ll_finish_md_op_data(op_data);
1126         if (!err) {
1127                 ll_update_times(request, src);
1128                 ll_update_times(request, tgt);
1129                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1130                 err = ll_objects_destroy(request, src);
1131         }
1132
1133         ptlrpc_req_finished(request);
1134
1135         RETURN(err);
1136 }
1137
1138 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1139                     dev_t rdev)
1140 {
1141         return ll_mknod_generic(dir, &dchild->d_name, mode,
1142                                 old_encode_dev(rdev), dchild);
1143 }
1144
1145 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1146 {
1147         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1148 }
1149
1150 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1151 {
1152         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1153 }
1154
1155 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1156 {
1157         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1158 }
1159
1160 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1161                       const char *oldname)
1162 {
1163         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1164 }
1165
1166 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1167                    struct dentry *new_dentry)
1168 {
1169         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1170                                new_dentry);
1171 }
1172
1173 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1174                      struct inode *new_dir, struct dentry *new_dentry)
1175 {
1176         int err;
1177         err = ll_rename_generic(old_dir, NULL,
1178                                  old_dentry, &old_dentry->d_name,
1179                                  new_dir, NULL, new_dentry,
1180                                  &new_dentry->d_name);
1181         if (!err) {
1182 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1183                 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1184 #endif
1185                         d_move(old_dentry, new_dentry);
1186         }
1187         return err;
1188 }
1189
1190 struct inode_operations ll_dir_inode_operations = {
1191         .mknod              = ll_mknod,
1192         .lookup             = ll_lookup_nd,
1193         .create             = ll_create_nd,
1194         /* We need all these non-raw things for NFSD, to not patch it. */
1195         .unlink             = ll_unlink,
1196         .mkdir              = ll_mkdir,
1197         .rmdir              = ll_rmdir,
1198         .symlink            = ll_symlink,
1199         .link               = ll_link,
1200         .rename             = ll_rename,
1201         .setattr            = ll_setattr,
1202         .getattr            = ll_getattr,
1203         .permission         = ll_inode_permission,
1204         .setxattr           = ll_setxattr,
1205         .getxattr           = ll_getxattr,
1206         .listxattr          = ll_listxattr,
1207         .removexattr        = ll_removexattr,
1208 #ifdef HAVE_IOP_GET_ACL
1209         .get_acl            = ll_get_acl,
1210 #endif
1211 };
1212
1213 struct inode_operations ll_special_inode_operations = {
1214         .setattr        = ll_setattr,
1215         .getattr        = ll_getattr,
1216         .permission     = ll_inode_permission,
1217         .setxattr       = ll_setxattr,
1218         .getxattr       = ll_getxattr,
1219         .listxattr      = ll_listxattr,
1220         .removexattr    = ll_removexattr,
1221 #ifdef HAVE_IOP_GET_ACL
1222         .get_acl            = ll_get_acl,
1223 #endif
1224 };