Whamcloud - gitweb
- fixes for dentry problems from Phil and myself
[fs/lustre-release.git] / lustre / llite / namei.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * This code is issued under the GNU General Public License.
5  * See the file COPYING in this distribution
6  *
7  * Copyright (C) 1992, 1993, 1994, 1995
8  * Remy Card (card@masi.ibp.fr)
9  * Laboratoire MASI - Institut Blaise Pascal
10  * Universite Pierre et Marie Curie (Paris VI)
11  *
12  *  from
13  *
14  *  linux/fs/ext2/namei.c
15  *
16  *  Copyright (C) 1991, 1992  Linus Torvalds
17  *
18  *  Big-endian to little-endian byte-swapping/bitmaps by
19  *        David S. Miller (davem@caip.rutgers.edu), 1995
20  *  Directory entry file type support and forward compatibility hooks
21  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
22  *
23  *  Changes for use in OBDFS
24  *  Copyright (c) 1999, Seagate Technology Inc.
25  *  Copyright (C) 2001, Cluster File Systems, Inc.
26  *                       Rewritten based on recent ext2 page cache use.
27  *
28  */
29
30 #include <linux/fs.h>
31 #include <linux/sched.h>
32 #include <linux/mm.h>
33 #include <linux/smp_lock.h>
34 #include <linux/quotaops.h>
35 #include <linux/highmem.h>
36 #include <linux/pagemap.h>
37
38 #define DEBUG_SUBSYSTEM S_LLITE
39
40 #include <linux/obd_support.h>
41 #include <linux/lustre_lite.h>
42 #include <linux/lustre_dlm.h>
43 #include <linux/obd_lov.h>
44
45 extern struct address_space_operations ll_aops;
46
47 /* from super.c */
48 extern void ll_change_inode(struct inode *inode);
49 extern int ll_setattr(struct dentry *de, struct iattr *attr);
50
51 /* from dir.c */
52 extern int ll_add_link (struct dentry *dentry, struct inode *inode);
53 obd_id ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *typ);
54 int ext2_make_empty(struct inode *inode, struct inode *parent);
55 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
56                    struct dentry *dentry, struct page ** res_page);
57 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page );
58 int ext2_empty_dir (struct inode * inode);
59 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p);
60 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
61                    struct page *page, struct inode *inode);
62
63 /*
64  * Couple of helper functions - make the code slightly cleaner.
65  */
66 static inline void ext2_inc_count(struct inode *inode)
67 {
68         inode->i_nlink++;
69 }
70
71 /* postpone the disk update until the inode really goes away */
72 static inline void ext2_dec_count(struct inode *inode)
73 {
74         inode->i_nlink--;
75 }
76
77 static inline int ext2_add_nondir(struct dentry *dentry, struct inode *inode)
78 {
79         int err;
80         err = ll_add_link(dentry, inode);
81         if (!err) {
82                 d_instantiate(dentry, inode);
83                 return 0;
84         }
85         ext2_dec_count(inode);
86         iput(inode);
87         return err;
88 }
89
90 /* methods */
91
92 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
93 static int ll_find_inode(struct inode *inode, unsigned long ino, void *opaque)
94 #else
95 static int ll_test_inode(struct inode *inode, void *opaque)
96 #endif
97 {
98         struct ll_read_inode2_cookie *lic = opaque;
99         struct mds_body *body = lic->lic_body;
100
101         if (inode->i_generation != lic->lic_body->generation)
102                 return 0;
103
104         /* Apply the attributes in 'opaque' to this inode */
105         ll_update_inode(inode, body);
106
107         return 1;
108 }
109
110 extern struct dentry_operations ll_d_ops;
111
112 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
113 {
114         ENTRY;
115
116         ldlm_lock_decref(lockh, mode);
117
118         RETURN(0);
119 }
120
121 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
122 extern int ll_read_inode2(struct inode *inode, void *opaque);
123 struct inode *ll_iget(struct super_block *sb, ino_t hash,
124                       struct ll_read_inode2_cookie *lic)
125 {
126         struct inode *inode;
127
128         inode = iget5_locked(sb, hash, ll_test_inode, ll_read_inode2, lic);
129
130         if (!inode)
131                 return ERR_PTR(-ENOMEM);
132
133         if (inode->i_state & I_NEW) {
134
135                 unlock_new_inode(inode);
136         }
137
138         // XXX Coda always fills inodes, should Lustre?
139         return inode;
140 }
141 #else
142 struct inode *ll_iget(struct super_block *sb, ino_t hash,
143                       struct ll_read_inode2_cookie *lic)
144 {
145         struct inode *inode;
146         inode = iget4(sb, hash, ll_find_inode, lic);
147         return inode;
148 }
149 #endif
150
151 static int ll_intent_to_lock_mode(struct lookup_intent *it)
152 {
153         /* CREAT needs to be tested before open (both could be set) */
154         if ((it->it_op & (IT_CREAT | IT_MKDIR | IT_SETATTR | IT_MKNOD))) {
155                 return LCK_PW;
156         } else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_UNLINK |
157                                 IT_RMDIR | IT_RENAME | IT_RENAME2 | IT_READLINK|
158                                 IT_LINK | IT_LINK2 | IT_LOOKUP | IT_SYMLINK)) {
159                 return LCK_PR;
160         }
161
162         LBUG();
163         RETURN(-EINVAL);
164 }
165
166 #define LL_LOOKUP_POSITIVE 1
167 #define LL_LOOKUP_NEGATIVE 2
168
169 int ll_intent_lock(struct inode *parent, struct dentry **de,
170                    struct lookup_intent *it,
171                    intent_finish_cb intent_finish)
172 {
173         struct dentry *dentry = *de;
174         struct ll_sb_info *sbi = ll_i2sbi(parent);
175         struct lustre_handle lockh;
176         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
177         struct ptlrpc_request *request = NULL;
178         char *tgt = NULL;
179         int rc, lock_mode, tgtlen = 0, offset, flag = LL_LOOKUP_POSITIVE;
180         obd_id ino = 0;
181
182         ENTRY;
183
184         if (it == NULL)
185                 it = &lookup_it;
186
187         CDEBUG(D_INFO, "name: %*s, intent: %s\n", dentry->d_name.len,
188                dentry->d_name.name, ldlm_it2str(it->it_op));
189
190         if (dentry->d_name.len > EXT2_NAME_LEN)
191                 RETURN(-ENAMETOOLONG);
192
193         lock_mode = ll_intent_to_lock_mode(it);
194         if (it->it_op & IT_SYMLINK) {
195                 tgt = it->it_data;
196                 tgtlen = strlen(tgt);
197                 it->it_data = NULL;
198         }
199
200         rc = mdc_enqueue(&sbi->ll_mdc_conn, LDLM_PLAIN, it, lock_mode, parent,
201                          dentry, &lockh, tgt, tgtlen, parent, sizeof(*parent));
202         if (rc < 0)
203                 RETURN(rc);
204         memcpy(it->it_lock_handle, &lockh, sizeof(lockh));
205
206         request = (struct ptlrpc_request *)it->it_data;
207         if (it->it_disposition) {
208                 struct mds_body *mds_body;
209                 int mode, symlen = 0;
210                 obd_flag valid;
211
212                 /* it_disposition == 1 indicates that the server performed the
213                  * intent on our behalf.  This long block is all about fixing up
214                  * the local state so that it is correct as of the moment
215                  * _before_ the operation was applied; that way, the VFS will
216                  * think that everything is normal and call Lustre's regular
217                  * FS function.
218                  *
219                  * If we're performing a creation, that means that unless the
220                  * creation failed with EEXIST, we should fake up a negative
221                  * dentry.  Likewise for the target of a hard link.
222                  *
223                  * For everything else, we want to lookup to succeed. */
224
225                 /* One additional note: we add an extra reference to the request
226                  * because we need to keep it around until ll_create gets
227                  * called.  For anything else which results in
228                  * LL_LOOKUP_POSITIVE, we can do the iget() immediately with the
229                  * contents of the reply (in the intent_finish callback).  In
230                  * the create case, however, we need to wait until
231                  * ll_create_node to do the iget() or the VFS will abort with
232                  * -EEXISTS. */
233
234                 offset = 1;
235                 mds_body = lustre_msg_buf(request->rq_repmsg, offset);
236                 ino = mds_body->fid1.id;
237                 mode = mds_body->mode;
238                 if (it->it_op & (IT_CREAT | IT_MKDIR | IT_SYMLINK | IT_MKNOD)) {
239                         mdc_store_create_replay_data(request, parent->i_sb);
240                         /* For create ops, we want the lookup to be negative,
241                          * unless the create failed in a way that indicates
242                          * that the file is already there */
243                         if (it->it_status != -EEXIST) {
244                                 atomic_inc(&request->rq_refcount);
245                                 GOTO(out, flag = LL_LOOKUP_NEGATIVE);
246                         }
247                         /* Fall through to update attibutes. */
248                 } else if (it->it_op & (IT_GETATTR | IT_SETATTR | IT_LOOKUP |
249                                         IT_READLINK)) {
250                         /* For check ops, we want the lookup to succeed */
251                         it->it_data = NULL;
252                         if (it->it_status)
253                                 GOTO(out, flag = LL_LOOKUP_NEGATIVE);
254                         /* Fall through to update attibutes. */
255                 } else if (it->it_op & (IT_RENAME | IT_LINK)) {
256                         /* For rename, we want the source lookup to succeed */
257                         if (it->it_status) {
258                                 it->it_data = NULL;
259                                 GOTO(drop_req, rc = it->it_status);
260                         }
261                         it->it_data = dentry;
262                         /* Fall through to update attibutes. */
263                 } else if (it->it_op & (IT_UNLINK | IT_RMDIR)) {
264                         /* For remove ops, we want the lookup to succeed unless
265                          * the file truly doesn't exist */
266                         it->it_data = NULL;
267                         if (it->it_status == -ENOENT)
268                                 GOTO(out, flag = LL_LOOKUP_NEGATIVE);
269                         /* No point in updating attributes that we're about to
270                          * unlink.  -phil */
271                         GOTO(out, flag = LL_LOOKUP_POSITIVE);
272                 } else if (it->it_op == IT_OPEN) {
273                         it->it_data = NULL;
274                         if (it->it_status && it->it_status != -EEXIST)
275                                 GOTO(out, flag = LL_LOOKUP_NEGATIVE);
276                         /* Fall through to update attibutes. */
277                 } else if (it->it_op & (IT_RENAME2 | IT_LINK2)) {
278                         it->it_data = NULL;
279                         /* This means the target lookup is negative */
280                         if (mds_body->valid == 0)
281                                 GOTO(out, flag = LL_LOOKUP_NEGATIVE);
282                         /* XXX bug 289: should we maybe fall through here? -p */
283                         GOTO(out, flag = LL_LOOKUP_POSITIVE);
284                 }
285
286                 /* Do a getattr now that we have the lock */
287                 valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
288                 if (it->it_op == IT_READLINK) {
289                         valid |= OBD_MD_LINKNAME;
290                         symlen = mds_body->size;
291                 }
292                 ptlrpc_req_finished(request);
293                 request = NULL;
294                 rc = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
295                                  valid, symlen, &request);
296                 if (rc) {
297                         CERROR("failure %d inode "LPX64"\n", rc, ino);
298                         GOTO(drop_req, rc = -abs(rc));
299                 }
300                 offset = 0;
301         } else {
302                 struct ll_inode_info *lli = ll_i2info(parent);
303                 int mode;
304
305                 /* it_disposition == 0 indicates that it just did a simple lock
306                  * request, for which we are very thankful.  move along with
307                  * the local lookup then. */
308
309                 memcpy(&lli->lli_intent_lock_handle, &lockh, sizeof(lockh));
310                 offset = 0;
311
312                 ino = ll_inode_by_name(parent, dentry, &mode);
313                 if (!ino) {
314                         CERROR("inode %*s not found by name\n",
315                                dentry->d_name.len, dentry->d_name.name);
316                         GOTO(drop_lock, rc = -ENOENT);
317                 }
318
319                 rc = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
320                                  OBD_MD_FLNOTOBD|OBD_MD_FLEASIZE, 0, &request);
321                 if (rc) {
322                         CERROR("failure %d inode "LPX64"\n", rc, ino);
323                         GOTO(drop_req, rc = -abs(rc));
324                 }
325         }
326
327         EXIT;
328  out:
329         if (intent_finish != NULL) {
330                 rc = intent_finish(flag, request, de, it, offset, ino);
331                 dentry = *de; /* intent_finish may change *de */
332         } else {
333                 ptlrpc_req_finished(request);
334         }
335
336         if (it->it_op == IT_LOOKUP || rc < 0)
337                 ll_intent_release(dentry, it);
338
339         return rc;
340
341  drop_req:
342         ptlrpc_free_req(request);
343  drop_lock:
344 #warning FIXME: must release lock here
345         return rc;
346 }
347
348 /* Search "inode"'s alias list for a dentry that has the same name and parent as
349  * de.  If found, return it.  If not found, return de. */
350 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
351 {
352         struct list_head *tmp;
353
354         spin_lock(&dcache_lock);
355         list_for_each(tmp, &inode->i_dentry) {
356                 struct dentry *dentry = list_entry(tmp, struct dentry, d_alias);
357
358                 /* We are called here with 'de' already on the aliases list. */
359                 if (dentry == de)
360                         continue;
361
362                 if (!atomic_read(&dentry->d_count))
363                         continue;
364
365                 if (!list_empty(&dentry->d_lru))
366                         continue;
367
368                 if (dentry->d_parent != de->d_parent)
369                         continue;
370
371                 if (dentry->d_name.len != de->d_name.len)
372                         continue;
373
374                 if (memcmp(dentry->d_name.name, de->d_name.name,
375                            de->d_name.len) != 0)
376                         continue;
377
378                 list_del_init(&dentry->d_hash);
379
380                 spin_unlock(&dcache_lock);
381                 d_rehash(dentry);
382                 atomic_inc(&dentry->d_count);
383                 iput(inode);
384                 return dentry;
385         }
386
387         spin_unlock(&dcache_lock);
388
389         return de;
390 }
391
392 static int
393 lookup2_finish(int flag, struct ptlrpc_request *request, struct dentry **de,
394                struct lookup_intent *it, int offset, obd_id ino)
395 {
396         struct dentry *dentry = *de, *saved = *de;
397         struct inode *inode = NULL;
398         struct ll_read_inode2_cookie lic;
399
400         if (flag == LL_LOOKUP_POSITIVE) {
401                 ENTRY;
402                 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
403
404                 if (S_ISREG(lic.lic_body->mode) &&
405                     lic.lic_body->valid & OBD_MD_FLEASIZE) {
406                         LASSERT(request->rq_repmsg->bufcount > offset);
407                         lic.lic_lmm = lustre_msg_buf(request->rq_repmsg,
408                                                      offset + 1);
409                 } else {
410                         lic.lic_lmm = NULL;
411                 }
412
413                 /* No rpc's happen during iget4, -ENOMEM's are possible */
414                 LASSERT(ino != 0);
415                 inode = ll_iget(dentry->d_sb, ino, &lic);
416                 if (!inode) {
417                         /* XXX make sure that request is freed in this case;
418                          * I think it is, but double-check refcounts. -phil */
419                         RETURN(-ENOMEM);
420                 }
421
422                 dentry = *de = ll_find_alias(inode, dentry);
423
424                 /* We asked for a lock on the directory, and may have been
425                  * granted a lock on the inode.  Just in case, fixup the data
426                  * pointer. */
427                 ldlm_lock_set_data((struct lustre_handle *)it->it_lock_handle,
428                                    inode, sizeof(*inode));
429
430                 EXIT;
431         } else {
432                 ENTRY;
433         }
434
435         ptlrpc_req_finished(request);
436
437         dentry->d_op = &ll_d_ops;
438         if (ll_d2d(dentry) == NULL) {
439                 ll_set_dd(dentry);
440         }
441
442         if (dentry == saved)
443                 d_add(dentry, inode);
444
445         if (it->it_status == 0 && it->it_op != IT_RENAME2) {
446                 LL_SAVE_INTENT(dentry, it);
447         } else {
448                 dentry->d_it = NULL;
449                 CDEBUG(D_DENTRY,
450                        "D_IT dentry %p fsdata %p intent: %s status %d\n",
451                        dentry, ll_d2d(dentry), ldlm_it2str(it->it_op),
452                        it->it_status);
453         }
454
455         RETURN(0);
456 }
457
458 static struct dentry *ll_lookup2(struct inode *parent, struct dentry *dentry,
459                                  struct lookup_intent *it)
460 {
461         struct dentry *save = dentry;
462         int rc;
463
464         rc = ll_intent_lock(parent, &dentry, it, lookup2_finish);
465         if (rc < 0) {
466                 CERROR("ll_intent_lock: %d\n", rc);
467                 return ERR_PTR(rc);
468         }
469
470         if (dentry == save)
471                 return NULL;
472         else
473                 return dentry;
474 }
475
476 static struct inode *ll_create_node(struct inode *dir, const char *name,
477                                     int namelen, const char *tgt, int tgtlen,
478                                     int mode, __u64 extra,
479                                     struct lookup_intent *it,
480                                     struct lov_stripe_md *lsm)
481 {
482         struct inode *inode;
483         struct ptlrpc_request *request = NULL;
484         struct mds_body *body;
485         time_t time = CURRENT_TIME;
486         struct ll_sb_info *sbi = ll_i2sbi(dir);
487         struct ll_read_inode2_cookie lic;
488         struct lov_mds_md *lmm = NULL;
489         ENTRY;
490
491         if (it && it->it_disposition) {
492                 int rc = it->it_status;
493                 if (rc) {
494                         CERROR("error creating MDS inode for %*s: rc = %d\n",
495                                namelen, name, rc);
496                         RETURN(ERR_PTR(rc));
497                 }
498                 ll_invalidate_inode_pages(dir);
499                 request = it->it_data;
500                 body = lustre_msg_buf(request->rq_repmsg, 1);
501                 lic.lic_lmm = NULL;
502         } else {
503                 int gid = current->fsgid;
504                 int rc;
505
506                 if (lsm) {
507                         OBD_ALLOC(lmm, lsm->lsm_mds_easize);
508                         if (!lmm)
509                                 RETURN(ERR_PTR(-ENOMEM));
510                         lov_packmd(lmm, lsm);
511                         lic.lic_lmm = lmm;
512                 } else
513                         lic.lic_lmm = NULL;
514
515                 if (dir->i_mode & S_ISGID) {
516                         gid = dir->i_gid;
517                         if (S_ISDIR(mode))
518                                 mode |= S_ISGID;
519                 }
520
521                 rc = mdc_create(&sbi->ll_mdc_conn, dir, name, namelen, tgt,
522                                 tgtlen, mode, current->fsuid, gid,
523                                 time, extra, lsm, &request);
524                 if (rc) {
525                         inode = ERR_PTR(rc);
526                         GOTO(out, rc);
527                 }
528                 body = lustre_msg_buf(request->rq_repmsg, 0);
529         }
530
531         lic.lic_body = body;
532
533         LASSERT(body->ino != 0);
534         inode = ll_iget(dir->i_sb, body->ino, &lic);
535         if (IS_ERR(inode)) {
536                 int rc = PTR_ERR(inode);
537                 CERROR("new_inode -fatal: rc %d\n", rc);
538                 LBUG();
539                 GOTO(out, rc);
540         }
541
542         if (!list_empty(&inode->i_dentry)) {
543                 CERROR("new_inode -fatal: inode %d, ct %d lnk %d\n",
544                        body->ino, atomic_read(&inode->i_count),
545                        inode->i_nlink);
546                 iput(inode);
547                 LBUG();
548                 inode = ERR_PTR(-EIO);
549                 GOTO(out, -EIO);
550         }
551
552         if (it && it->it_disposition) {
553                 /* We asked for a lock on the directory, but were
554                  * granted a lock on the inode.  Since we finally have
555                  * an inode pointer, stuff it in the lock. */
556                 ldlm_lock_set_data((struct lustre_handle *)it->it_lock_handle,
557                                    inode, sizeof(*inode));
558         }
559
560         EXIT;
561  out:
562         if (lsm && lmm)
563                 OBD_FREE(lmm, lsm->lsm_mds_easize);
564         ptlrpc_req_finished(request);
565         return inode;
566 }
567
568 static int ll_mdc_unlink(struct inode *dir, struct inode *child, __u32 mode,
569                          const char *name, int len)
570 {
571         struct ptlrpc_request *request = NULL;
572         struct ll_sb_info *sbi = ll_i2sbi(dir);
573         int err;
574
575         ENTRY;
576
577         err = mdc_unlink(&sbi->ll_mdc_conn, dir, child, mode, name, len,
578                          &request);
579         ptlrpc_req_finished(request);
580
581         RETURN(err);
582 }
583
584 int ll_mdc_link(struct dentry *src, struct inode *dir,
585                 const char *name, int len)
586 {
587         struct ptlrpc_request *request = NULL;
588         int err;
589         struct ll_sb_info *sbi = ll_i2sbi(dir);
590
591         ENTRY;
592
593         err = mdc_link(&sbi->ll_mdc_conn, src, dir, name, len, &request);
594         ptlrpc_req_finished(request);
595
596         RETURN(err);
597 }
598
599 int ll_mdc_rename(struct inode *src, struct inode *tgt,
600                   struct dentry *old, struct dentry *new)
601 {
602         struct ptlrpc_request *request = NULL;
603         struct ll_sb_info *sbi = ll_i2sbi(src);
604         int err;
605
606         ENTRY;
607
608         err = mdc_rename(&sbi->ll_mdc_conn, src, tgt,
609                          old->d_name.name, old->d_name.len,
610                          new->d_name.name, new->d_name.len, &request);
611         ptlrpc_req_finished(request);
612
613         RETURN(err);
614 }
615
616 /*
617  * By the time this is called, we already have created the directory cache
618  * entry for the new file, but it is so far negative - it has no inode.
619  * We defer creating the OBD object(s) until open, to keep the intent and
620  * non-intent code paths similar, and also because we do not have the MDS
621  * inode number before calling ll_create_node() (which is needed for LOV),
622  * so we would need to do yet another RPC to the MDS to store the LOV EA
623  * data on the MDS.
624  *
625  * If the create succeeds, we fill in the inode information
626  * with d_instantiate().
627  */
628 static int ll_create(struct inode *dir, struct dentry *dentry, int mode)
629 {
630         struct lookup_intent *it;
631         struct inode *inode;
632         int rc = 0;
633         ENTRY;
634
635         LL_GET_INTENT(dentry, it);
636
637         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
638                                NULL, 0, mode, 0, it, NULL);
639
640         if (IS_ERR(inode))
641                 RETURN(PTR_ERR(inode));
642
643         if (it->it_disposition) {
644                 struct ll_inode_info *lli = ll_i2info(inode);
645                 memcpy(&lli->lli_intent_lock_handle, it->it_lock_handle,
646                        sizeof(lli->lli_intent_lock_handle));
647                 d_instantiate(dentry, inode);
648         } else {
649                 /* no directory data updates when intents rule */
650                 rc = ext2_add_nondir(dentry, inode);
651         }
652
653         RETURN(rc);
654 }
655
656 static int ll_mknod(struct inode *dir, struct dentry *dentry, int mode,
657                     int rdev)
658 {
659         struct lookup_intent *it;
660         struct inode *inode;
661         int rc = 0;
662
663         LL_GET_INTENT(dentry, it);
664
665         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
666                                NULL, 0, mode, rdev, it, NULL);
667
668         if (IS_ERR(inode))
669                 RETURN(PTR_ERR(inode));
670
671         /* no directory data updates when intents rule */
672         if (it && it->it_disposition)
673                 d_instantiate(dentry, inode);
674         else
675                 rc = ext2_add_nondir(dentry, inode);
676
677         return rc;
678 }
679
680 static int ll_symlink(struct inode *dir, struct dentry *dentry,
681                       const char *symname)
682 {
683         struct lookup_intent *it;
684         unsigned l = strlen(symname);
685         struct inode *inode;
686         struct ll_inode_info *lli;
687         int err = 0;
688         ENTRY;
689
690         LL_GET_INTENT(dentry, it);
691
692         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
693                                symname, l, S_IFLNK | S_IRWXUGO, 0, it, NULL);
694         if (IS_ERR(inode))
695                 RETURN(PTR_ERR(inode));
696
697         lli = ll_i2info(inode);
698
699         OBD_ALLOC(lli->lli_symlink_name, l + 1);
700         /* this _could_ be a non-fatal error, since the symlink is already
701          * stored on the MDS by this point, and we can re-get it in readlink.
702          */
703         if (!lli->lli_symlink_name)
704                 RETURN(-ENOMEM);
705
706         memcpy(lli->lli_symlink_name, symname, l + 1);
707         inode->i_size = l;
708
709         /* no directory data updates when intents rule */
710         if (it && it->it_disposition)
711                 d_instantiate(dentry, inode);
712         else
713                 err = ext2_add_nondir(dentry, inode);
714
715         RETURN(err);
716 }
717
718 static int ll_link(struct dentry *old_dentry, struct inode * dir,
719                    struct dentry *dentry)
720 {
721         struct lookup_intent *it;
722         struct inode *inode = old_dentry->d_inode;
723         int rc;
724
725         LL_GET_INTENT(dentry, it);
726
727         if (it && it->it_disposition) {
728                 if (it->it_status)
729                         RETURN(it->it_status);
730                 inode->i_ctime = CURRENT_TIME;
731                 ext2_inc_count(inode);
732                 atomic_inc(&inode->i_count);
733                 d_instantiate(dentry, inode);
734                 ll_invalidate_inode_pages(dir);
735                 RETURN(0);
736         }
737
738         if (S_ISDIR(inode->i_mode))
739                 return -EPERM;
740
741         if (inode->i_nlink >= EXT2_LINK_MAX)
742                 return -EMLINK;
743
744         rc = ll_mdc_link(old_dentry, dir,
745                           dentry->d_name.name, dentry->d_name.len);
746         if (rc)
747                 RETURN(rc);
748
749         inode->i_ctime = CURRENT_TIME;
750         ext2_inc_count(inode);
751         atomic_inc(&inode->i_count);
752
753         return ext2_add_nondir(dentry, inode);
754 }
755
756 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
757 {
758         struct lookup_intent *it;
759         struct inode * inode;
760         int err = -EMLINK;
761         ENTRY;
762
763         LL_GET_INTENT(dentry, it);
764
765         if (dir->i_nlink >= EXT2_LINK_MAX)
766                 goto out;
767
768         ext2_inc_count(dir);
769         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
770                                NULL, 0, S_IFDIR | mode, 0, it, NULL);
771         err = PTR_ERR(inode);
772         if (IS_ERR(inode))
773                 goto out_dir;
774
775         ext2_inc_count(inode);
776
777         err = ext2_make_empty(inode, dir);
778         if (err)
779                 goto out_fail;
780
781         /* no directory data updates when intents rule */
782         if (!it || !it->it_disposition) {
783                 err = ll_add_link(dentry, inode);
784                 if (err)
785                         goto out_fail;
786         }
787
788         d_instantiate(dentry, inode);
789 out:
790         EXIT;
791         return err;
792
793 out_fail:
794         ext2_dec_count(inode);
795         ext2_dec_count(inode);
796         iput(inode);
797         EXIT;
798 out_dir:
799         ext2_dec_count(dir);
800         EXIT;
801         goto out;
802 }
803
804 static int ll_common_unlink(struct inode *dir, struct dentry *dentry,
805                             struct lookup_intent *it, __u32 mode)
806 {
807         struct inode *inode = dentry->d_inode;
808         struct ext2_dir_entry_2 * de;
809         struct page * page;
810         int rc = 0;
811
812         if (it && it->it_disposition) {
813                 rc = it->it_status;
814                 ll_invalidate_inode_pages(dir);
815                 if (rc)
816                         GOTO(out, rc);
817                 GOTO(out_dec, 0);
818         }
819
820         de = ext2_find_entry(dir, dentry, &page);
821         if (!de)
822                 GOTO(out, rc = -ENOENT);
823         rc = ll_mdc_unlink(dir, dentry->d_inode, mode,
824                            dentry->d_name.name, dentry->d_name.len);
825         if (rc)
826                 GOTO(out, rc);
827
828         rc = ext2_delete_entry(de, page);
829         if (rc)
830                 GOTO(out, rc);
831
832         /* AED: not sure if needed - directory lock revocation should do it
833          * in the case where the client has cached it for non-intent ops.
834          */
835         ll_invalidate_inode_pages(dir);
836
837         inode->i_ctime = dir->i_ctime;
838 out_dec:
839         ext2_dec_count(inode);
840 out:
841         return rc;
842 }
843
844 static int ll_unlink(struct inode *dir, struct dentry *dentry)
845 {
846         struct lookup_intent * it;
847
848         LL_GET_INTENT(dentry, it);
849
850         return ll_common_unlink(dir, dentry, it, S_IFREG);
851 }
852
853 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
854 {
855         struct inode * inode = dentry->d_inode;
856         struct lookup_intent *it;
857         int rc;
858         ENTRY;
859
860         LL_GET_INTENT(dentry, it);
861
862         if ((!it || !it->it_disposition) && !ext2_empty_dir(inode))
863                 RETURN(-ENOTEMPTY);
864
865         rc = ll_common_unlink(dir, dentry, it, S_IFDIR);
866         if (!rc) {
867                 inode->i_size = 0;
868                 ext2_dec_count(inode);
869                 ext2_dec_count(dir);
870         }
871
872         RETURN(rc);
873 }
874
875 static int ll_rename(struct inode * old_dir, struct dentry * old_dentry,
876                      struct inode * new_dir, struct dentry * new_dentry)
877 {
878         struct lookup_intent *it;
879         struct inode * old_inode = old_dentry->d_inode;
880         struct inode * tgt_inode = new_dentry->d_inode;
881         struct page * dir_page = NULL;
882         struct ext2_dir_entry_2 * dir_de = NULL;
883         struct ext2_dir_entry_2 * old_de;
884         struct page * old_page;
885         int err;
886
887         LL_GET_INTENT(old_dentry, it);
888
889         if (it && it->it_disposition) {
890                 if (tgt_inode) {
891                         tgt_inode->i_ctime = CURRENT_TIME;
892                         tgt_inode->i_nlink--;
893                 }
894                 ll_invalidate_inode_pages(old_dir);
895                 ll_invalidate_inode_pages(new_dir);
896                 GOTO(out, err = it->it_status);
897         }
898
899         err = ll_mdc_rename(old_dir, new_dir, old_dentry, new_dentry);
900         if (err)
901                 goto out;
902
903         old_de = ext2_find_entry (old_dir, old_dentry, &old_page);
904         if (!old_de)
905                 goto out;
906
907         if (S_ISDIR(old_inode->i_mode)) {
908                 err = -EIO;
909                 dir_de = ext2_dotdot(old_inode, &dir_page);
910                 if (!dir_de)
911                         goto out_old;
912         }
913
914         if (tgt_inode) {
915                 struct page *new_page;
916                 struct ext2_dir_entry_2 *new_de;
917
918                 err = -ENOTEMPTY;
919                 if (dir_de && !ext2_empty_dir (tgt_inode))
920                         goto out_dir;
921
922                 err = -ENOENT;
923                 new_de = ext2_find_entry (new_dir, new_dentry, &new_page);
924                 if (!new_de)
925                         goto out_dir;
926                 ext2_inc_count(old_inode);
927                 ext2_set_link(new_dir, new_de, new_page, old_inode);
928                 tgt_inode->i_ctime = CURRENT_TIME;
929                 if (dir_de)
930                         tgt_inode->i_nlink--;
931                 ext2_dec_count(tgt_inode);
932         } else {
933                 if (dir_de) {
934                         err = -EMLINK;
935                         if (new_dir->i_nlink >= EXT2_LINK_MAX)
936                                 goto out_dir;
937                 }
938                 ext2_inc_count(old_inode);
939                 err = ll_add_link(new_dentry, old_inode);
940                 if (err) {
941                         ext2_dec_count(old_inode);
942                         goto out_dir;
943                 }
944                 if (dir_de)
945                         ext2_inc_count(new_dir);
946         }
947
948         ext2_delete_entry (old_de, old_page);
949         ext2_dec_count(old_inode);
950
951         if (dir_de) {
952                 ext2_set_link(old_inode, dir_de, dir_page, new_dir);
953                 ext2_dec_count(old_dir);
954         }
955         return 0;
956
957 out_dir:
958         if (dir_de) {
959                 kunmap(dir_page);
960                 page_cache_release(dir_page);
961         }
962 out_old:
963         kunmap(old_page);
964         page_cache_release(old_page);
965 out:
966         return err;
967 }
968
969 struct inode_operations ll_dir_inode_operations = {
970         create:         ll_create,
971         lookup2:        ll_lookup2,
972         link:           ll_link,
973         unlink:         ll_unlink,
974         symlink:        ll_symlink,
975         mkdir:          ll_mkdir,
976         rmdir:          ll_rmdir,
977         mknod:          ll_mknod,
978         rename:         ll_rename,
979         setattr:        ll_setattr
980 };