projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'staging-linus' into staging-work
[deliverable/linux.git] / drivers / staging / lustre / 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, Intel Corporation.
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 static int ll_create_it(struct inode *, struct dentry *,
55 int, struct lookup_intent *);
56
57 /*
58 * Check if we have something mounted at the named dchild.
59 * In such a case there would always be dentry present.
60 */
61 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
62 struct qstr *name)
63 {
64 int mounted = 0;
65
66 if (unlikely(dchild)) {
67 mounted = d_mountpoint(dchild);
68 } else if (dparent) {
69 dchild = d_lookup(dparent, name);
70 if (dchild) {
71 mounted = d_mountpoint(dchild);
72 dput(dchild);
73 }
74 }
75 return mounted;
76 }
77
78 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
79 {
80 ldlm_lock_decref(lockh, mode);
81
82 return 0;
83 }
84
85
86 /* called from iget5_locked->find_inode() under inode_hash_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
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 CDEBUG(D_INODE,
152 "%s: apply lsm %p to inode "DFID".\n",
153 ll_get_fsname(sb, NULL, 0), md->lsm,
154 PFID(ll_inode2fid(inode)));
155 rc = cl_file_inode_init(inode, md);
156 }
157 if (rc != 0) {
158 make_bad_inode(inode);
159 unlock_new_inode(inode);
160 iput(inode);
161 inode = ERR_PTR(rc);
162 } else
163 unlock_new_inode(inode);
164 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
165 ll_update_inode(inode, md);
166 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
167 inode, PFID(&md->body->fid1));
168 }
169 return inode;
170 }
171
172 static void ll_invalidate_negative_children(struct inode *dir)
173 {
174 struct dentry *dentry, *tmp_subdir;
175 struct ll_d_hlist_node *p;
176
177 ll_lock_dcache(dir);
178 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
179 spin_lock(&dentry->d_lock);
180 if (!list_empty(&dentry->d_subdirs)) {
181 struct dentry *child;
182
183 list_for_each_entry_safe(child, tmp_subdir,
184 &dentry->d_subdirs,
185 d_u.d_child) {
186 if (child->d_inode == NULL)
187 d_lustre_invalidate(child, 1);
188 }
189 }
190 spin_unlock(&dentry->d_lock);
191 }
192 ll_unlock_dcache(dir);
193 }
194
195 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
196 void *data, int flag)
197 {
198 int rc;
199 struct lustre_handle lockh;
200
201 switch (flag) {
202 case LDLM_CB_BLOCKING:
203 ldlm_lock2handle(lock, &lockh);
204 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
205 if (rc < 0) {
206 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
207 return rc;
208 }
209 break;
210 case LDLM_CB_CANCELING: {
211 struct inode *inode = ll_inode_from_resource_lock(lock);
212 struct ll_inode_info *lli;
213 __u64 bits = lock->l_policy_data.l_inodebits.bits;
214 struct lu_fid *fid;
215 ldlm_mode_t mode = lock->l_req_mode;
216
217 /* Inode is set to lock->l_resource->lr_lvb_inode
218 * for mdc - bug 24555 */
219 LASSERT(lock->l_ast_data == NULL);
220
221 /* Invalidate all dentries associated with this inode */
222 if (inode == NULL)
223 break;
224
225 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
226
227 if (bits & MDS_INODELOCK_XATTR)
228 ll_xattr_cache_destroy(inode);
229
230 /* For OPEN locks we differentiate between lock modes
231 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
232 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
233 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
234 ll_have_md_lock(inode, &bits, LCK_MINMODE);
235
236 if (bits & MDS_INODELOCK_OPEN)
237 ll_have_md_lock(inode, &bits, mode);
238
239 fid = ll_inode2fid(inode);
240 if (!fid_res_name_eq(fid, &lock->l_resource->lr_name))
241 LDLM_ERROR(lock, "data mismatch with object "
242 DFID" (%p)", PFID(fid), inode);
243
244 if (bits & MDS_INODELOCK_OPEN) {
245 int flags = 0;
246 switch (lock->l_req_mode) {
247 case LCK_CW:
248 flags = FMODE_WRITE;
249 break;
250 case LCK_PR:
251 flags = FMODE_EXEC;
252 break;
253 case LCK_CR:
254 flags = FMODE_READ;
255 break;
256 default:
257 CERROR("Unexpected lock mode for OPEN lock "
258 "%d, inode %ld\n", lock->l_req_mode,
259 inode->i_ino);
260 }
261 ll_md_real_close(inode, flags);
262 }
263
264 lli = ll_i2info(inode);
265 if (bits & MDS_INODELOCK_LAYOUT) {
266 struct cl_object_conf conf = { { 0 } };
267
268 conf.coc_opc = OBJECT_CONF_INVALIDATE;
269 conf.coc_inode = inode;
270 rc = ll_layout_conf(inode, &conf);
271 if (rc)
272 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
273 }
274
275 if (bits & MDS_INODELOCK_UPDATE) {
276 spin_lock(&lli->lli_lock);
277 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
278 spin_unlock(&lli->lli_lock);
279 }
280
281 if (S_ISDIR(inode->i_mode) &&
282 (bits & MDS_INODELOCK_UPDATE)) {
283 CDEBUG(D_INODE, "invalidating inode %lu\n",
284 inode->i_ino);
285 truncate_inode_pages(inode->i_mapping, 0);
286 ll_invalidate_negative_children(inode);
287 }
288
289 if (inode->i_sb->s_root &&
290 inode != inode->i_sb->s_root->d_inode &&
291 (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)))
292 ll_invalidate_aliases(inode);
293 iput(inode);
294 break;
295 }
296 default:
297 LBUG();
298 }
299
300 return 0;
301 }
302
303 __u32 ll_i2suppgid(struct inode *i)
304 {
305 if (in_group_p(i->i_gid))
306 return (__u32)from_kgid(&init_user_ns, i->i_gid);
307 else
308 return (__u32)(-1);
309 }
310
311 /* Pack the required supplementary groups into the supplied groups array.
312 * If we don't need to use the groups from the target inode(s) then we
313 * instead pack one or more groups from the user's supplementary group
314 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
315 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
316 {
317 #if 0
318 int i;
319 #endif
320
321 LASSERT(i1 != NULL);
322 LASSERT(suppgids != NULL);
323
324 suppgids[0] = ll_i2suppgid(i1);
325
326 if (i2)
327 suppgids[1] = ll_i2suppgid(i2);
328 else
329 suppgids[1] = -1;
330
331 #if 0
332 for (i = 0; i < current_ngroups; i++) {
333 if (suppgids[0] == -1) {
334 if (current_groups[i] != suppgids[1])
335 suppgids[0] = current_groups[i];
336 continue;
337 }
338 if (suppgids[1] == -1) {
339 if (current_groups[i] != suppgids[0])
340 suppgids[1] = current_groups[i];
341 continue;
342 }
343 break;
344 }
345 #endif
346 }
347
348 /*
349 * try to reuse three types of dentry:
350 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
351 * by concurrent .revalidate).
352 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
353 * be cleared by others calling d_lustre_revalidate).
354 * 3. DISCONNECTED alias.
355 */
356 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
357 {
358 struct dentry *alias, *discon_alias, *invalid_alias;
359 struct ll_d_hlist_node *p;
360
361 if (ll_d_hlist_empty(&inode->i_dentry))
362 return NULL;
363
364 discon_alias = invalid_alias = NULL;
365
366 ll_lock_dcache(inode);
367 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
368 LASSERT(alias != dentry);
369
370 spin_lock(&alias->d_lock);
371 if (alias->d_flags & DCACHE_DISCONNECTED)
372 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
373 discon_alias = alias;
374 else if (alias->d_parent == dentry->d_parent &&
375 alias->d_name.hash == dentry->d_name.hash &&
376 alias->d_name.len == dentry->d_name.len &&
377 memcmp(alias->d_name.name, dentry->d_name.name,
378 dentry->d_name.len) == 0)
379 invalid_alias = alias;
380 spin_unlock(&alias->d_lock);
381
382 if (invalid_alias)
383 break;
384 }
385 alias = invalid_alias ?: discon_alias ?: NULL;
386 if (alias) {
387 spin_lock(&alias->d_lock);
388 dget_dlock(alias);
389 spin_unlock(&alias->d_lock);
390 }
391 ll_unlock_dcache(inode);
392
393 return alias;
394 }
395
396 /*
397 * Similar to d_splice_alias(), but lustre treats invalid alias
398 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
399 */
400 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
401 {
402 struct dentry *new;
403 int rc;
404
405 if (inode) {
406 new = ll_find_alias(inode, de);
407 if (new) {
408 rc = ll_d_init(new);
409 if (rc < 0) {
410 dput(new);
411 return ERR_PTR(rc);
412 }
413 d_move(new, de);
414 iput(inode);
415 CDEBUG(D_DENTRY,
416 "Reuse dentry %p inode %p refc %d flags %#x\n",
417 new, new->d_inode, d_count(new), new->d_flags);
418 return new;
419 }
420 }
421 rc = ll_d_init(de);
422 if (rc < 0)
423 return ERR_PTR(rc);
424 d_add(de, inode);
425 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
426 de, de->d_inode, d_count(de), de->d_flags);
427 return de;
428 }
429
430 int ll_lookup_it_finish(struct ptlrpc_request *request,
431 struct lookup_intent *it, void *data)
432 {
433 struct it_cb_data *icbd = data;
434 struct dentry **de = icbd->icbd_childp;
435 struct inode *parent = icbd->icbd_parent;
436 struct inode *inode = NULL;
437 __u64 bits = 0;
438 int rc;
439
440 /* NB 1 request reference will be taken away by ll_intent_lock()
441 * when I return */
442 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
443 it->d.lustre.it_disposition);
444 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
445 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
446 if (rc)
447 return rc;
448
449 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
450
451 /* We used to query real size from OSTs here, but actually
452 this is not needed. For stat() calls size would be updated
453 from subsequent do_revalidate()->ll_inode_revalidate_it() in
454 2.4 and
455 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
456 Everybody else who needs correct file size would call
457 ll_glimpse_size or some equivalent themselves anyway.
458 Also see bug 7198. */
459 }
460
461 /* Only hash *de if it is unhashed (new dentry).
462 * Atoimc_open may passin hashed dentries for open.
463 */
464 if (d_unhashed(*de)) {
465 *de = ll_splice_alias(inode, *de);
466 if (IS_ERR(*de))
467 return PTR_ERR(*de);
468 }
469
470 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
471 /* we have lookup look - unhide dentry */
472 if (bits & MDS_INODELOCK_LOOKUP)
473 d_lustre_revalidate(*de);
474 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
475 /* If file created on server, don't depend on parent UPDATE
476 * lock to unhide it. It is left hidden and next lookup can
477 * find it in ll_splice_alias.
478 */
479 /* Check that parent has UPDATE lock. */
480 struct lookup_intent parent_it = {
481 .it_op = IT_GETATTR,
482 .d.lustre.it_lock_handle = 0 };
483
484 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
485 &ll_i2info(parent)->lli_fid, NULL)) {
486 d_lustre_revalidate(*de);
487 ll_intent_release(&parent_it);
488 }
489 }
490
491 return 0;
492 }
493
494 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
495 struct lookup_intent *it, int lookup_flags)
496 {
497 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
498 struct dentry *save = dentry, *retval;
499 struct ptlrpc_request *req = NULL;
500 struct md_op_data *op_data;
501 struct it_cb_data icbd;
502 __u32 opc;
503 int rc;
504
505 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
506 return ERR_PTR(-ENAMETOOLONG);
507
508 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
509 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
510 parent->i_generation, parent, LL_IT2STR(it));
511
512 if (d_mountpoint(dentry))
513 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
514
515 ll_frob_intent(&it, &lookup_it);
516
517 if (it->it_op == IT_GETATTR) {
518 rc = ll_statahead_enter(parent, &dentry, 0);
519 if (rc == 1) {
520 if (dentry == save)
521 GOTO(out, retval = NULL);
522 GOTO(out, retval = dentry);
523 }
524 }
525
526 icbd.icbd_childp = &dentry;
527 icbd.icbd_parent = parent;
528
529 if (it->it_op & IT_CREAT)
530 opc = LUSTRE_OPC_CREATE;
531 else
532 opc = LUSTRE_OPC_ANY;
533
534 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
535 dentry->d_name.len, lookup_flags, opc,
536 NULL);
537 if (IS_ERR(op_data))
538 return (void *)op_data;
539
540 /* enforce umask if acl disabled or MDS doesn't support umask */
541 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
542 it->it_create_mode &= ~current_umask();
543
544 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
545 lookup_flags, &req, ll_md_blocking_ast, 0);
546 ll_finish_md_op_data(op_data);
547 if (rc < 0)
548 GOTO(out, retval = ERR_PTR(rc));
549
550 rc = ll_lookup_it_finish(req, it, &icbd);
551 if (rc != 0) {
552 ll_intent_release(it);
553 GOTO(out, retval = ERR_PTR(rc));
554 }
555
556 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
557 !S_ISREG(dentry->d_inode->i_mode) &&
558 !S_ISDIR(dentry->d_inode->i_mode)) {
559 ll_release_openhandle(dentry, it);
560 }
561 ll_lookup_finish_locks(it, dentry);
562
563 if (dentry == save)
564 GOTO(out, retval = NULL);
565 else
566 GOTO(out, retval = dentry);
567 out:
568 if (req)
569 ptlrpc_req_finished(req);
570 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
571 ll_statahead_mark(parent, dentry);
572 return retval;
573 }
574
575 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
576 unsigned int flags)
577 {
578 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
579 struct dentry *de;
580
581 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),flags=%u\n",
582 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
583 parent->i_generation, parent, flags);
584
585 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
586 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
587 return NULL;
588
589 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
590 itp = NULL;
591 else
592 itp = &it;
593 de = ll_lookup_it(parent, dentry, itp, 0);
594
595 if (itp != NULL)
596 ll_intent_release(itp);
597
598 return de;
599 }
600
601 /*
602 * For cached negative dentry and new dentry, handle lookup/create/open
603 * together.
604 */
605 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
606 struct file *file, unsigned open_flags,
607 umode_t mode, int *opened)
608 {
609 struct lookup_intent *it;
610 struct dentry *de;
611 long long lookup_flags = LOOKUP_OPEN;
612 int rc = 0;
613
614 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),file %p,"
615 "open_flags %x,mode %x opened %d\n",
616 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
617 dir->i_generation, dir, file, open_flags, mode, *opened);
618
619 OBD_ALLOC(it, sizeof(*it));
620 if (!it)
621 return -ENOMEM;
622
623 it->it_op = IT_OPEN;
624 if (open_flags & O_CREAT) {
625 it->it_op |= IT_CREAT;
626 lookup_flags |= LOOKUP_CREATE;
627 }
628 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
629 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
630
631 /* Dentry added to dcache tree in ll_lookup_it */
632 de = ll_lookup_it(dir, dentry, it, lookup_flags);
633 if (IS_ERR(de))
634 rc = PTR_ERR(de);
635 else if (de != NULL)
636 dentry = de;
637
638 if (!rc) {
639 if (it_disposition(it, DISP_OPEN_CREATE)) {
640 /* Dentry instantiated in ll_create_it. */
641 rc = ll_create_it(dir, dentry, mode, it);
642 if (rc) {
643 /* We dget in ll_splice_alias. */
644 if (de != NULL)
645 dput(de);
646 goto out_release;
647 }
648
649 *opened |= FILE_CREATED;
650 }
651 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
652 /* Open dentry. */
653 if (S_ISFIFO(dentry->d_inode->i_mode)) {
654 /* We cannot call open here as it would
655 * deadlock.
656 */
657 if (it_disposition(it, DISP_ENQ_OPEN_REF))
658 ptlrpc_req_finished(
659 (struct ptlrpc_request *)
660 it->d.lustre.it_data);
661 rc = finish_no_open(file, de);
662 } else {
663 file->private_data = it;
664 rc = finish_open(file, dentry, NULL, opened);
665 /* We dget in ll_splice_alias. finish_open takes
666 * care of dget for fd open.
667 */
668 if (de != NULL)
669 dput(de);
670 }
671 } else {
672 rc = finish_no_open(file, de);
673 }
674 }
675
676 out_release:
677 ll_intent_release(it);
678 OBD_FREE(it, sizeof(*it));
679
680 return rc;
681 }
682
683
684 /* We depend on "mode" being set with the proper file type/umask by now */
685 static struct inode *ll_create_node(struct inode *dir, const char *name,
686 int namelen, const void *data, int datalen,
687 int mode, __u64 extra,
688 struct lookup_intent *it)
689 {
690 struct inode *inode = NULL;
691 struct ptlrpc_request *request = NULL;
692 struct ll_sb_info *sbi = ll_i2sbi(dir);
693 int rc;
694
695 LASSERT(it && it->d.lustre.it_disposition);
696
697 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
698 request = it->d.lustre.it_data;
699 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
700 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
701 if (rc)
702 GOTO(out, inode = ERR_PTR(rc));
703
704 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
705
706 /* We asked for a lock on the directory, but were granted a
707 * lock on the inode. Since we finally have an inode pointer,
708 * stuff it in the lock. */
709 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
710 inode, inode->i_ino, inode->i_generation);
711 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
712 out:
713 ptlrpc_req_finished(request);
714 return inode;
715 }
716
717 /*
718 * By the time this is called, we already have created the directory cache
719 * entry for the new file, but it is so far negative - it has no inode.
720 *
721 * We defer creating the OBD object(s) until open, to keep the intent and
722 * non-intent code paths similar, and also because we do not have the MDS
723 * inode number before calling ll_create_node() (which is needed for LOV),
724 * so we would need to do yet another RPC to the MDS to store the LOV EA
725 * data on the MDS. If needed, we would pass the PACKED lmm as data and
726 * lmm_size in datalen (the MDS still has code which will handle that).
727 *
728 * If the create succeeds, we fill in the inode information
729 * with d_instantiate().
730 */
731 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
732 struct lookup_intent *it)
733 {
734 struct inode *inode;
735 int rc = 0;
736
737 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
738 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
739 dir->i_generation, dir, LL_IT2STR(it));
740
741 rc = it_open_error(DISP_OPEN_CREATE, it);
742 if (rc)
743 return rc;
744
745 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
746 NULL, 0, mode, 0, it);
747 if (IS_ERR(inode))
748 return PTR_ERR(inode);
749
750 if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
751 ll_i2info(inode)->lli_volatile = true;
752
753 d_instantiate(dentry, inode);
754 return 0;
755 }
756
757 static void ll_update_times(struct ptlrpc_request *request,
758 struct inode *inode)
759 {
760 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
761 &RMF_MDT_BODY);
762
763 LASSERT(body);
764 if (body->valid & OBD_MD_FLMTIME &&
765 body->mtime > LTIME_S(inode->i_mtime)) {
766 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
767 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
768 LTIME_S(inode->i_mtime) = body->mtime;
769 }
770 if (body->valid & OBD_MD_FLCTIME &&
771 body->ctime > LTIME_S(inode->i_ctime))
772 LTIME_S(inode->i_ctime) = body->ctime;
773 }
774
775 static int ll_new_node(struct inode *dir, struct qstr *name,
776 const char *tgt, int mode, int rdev,
777 struct dentry *dchild, __u32 opc)
778 {
779 struct ptlrpc_request *request = NULL;
780 struct md_op_data *op_data;
781 struct inode *inode = NULL;
782 struct ll_sb_info *sbi = ll_i2sbi(dir);
783 int tgt_len = 0;
784 int err;
785
786 if (unlikely(tgt != NULL))
787 tgt_len = strlen(tgt) + 1;
788
789 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
790 name->len, 0, opc, NULL);
791 if (IS_ERR(op_data))
792 GOTO(err_exit, err = PTR_ERR(op_data));
793
794 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
795 from_kuid(&init_user_ns, current_fsuid()),
796 from_kgid(&init_user_ns, current_fsgid()),
797 cfs_curproc_cap_pack(), rdev, &request);
798 ll_finish_md_op_data(op_data);
799 if (err)
800 GOTO(err_exit, err);
801
802 ll_update_times(request, dir);
803
804 if (dchild) {
805 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
806 if (err)
807 GOTO(err_exit, err);
808
809 d_instantiate(dchild, inode);
810 }
811 err_exit:
812 ptlrpc_req_finished(request);
813
814 return err;
815 }
816
817 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
818 unsigned rdev, struct dentry *dchild)
819 {
820 int err;
821
822 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
823 name->len, name->name, dir->i_ino, dir->i_generation, dir,
824 mode, rdev);
825
826 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
827 mode &= ~current_umask();
828
829 switch (mode & S_IFMT) {
830 case 0:
831 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
832 case S_IFREG:
833 case S_IFCHR:
834 case S_IFBLK:
835 case S_IFIFO:
836 case S_IFSOCK:
837 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
838 LUSTRE_OPC_MKNOD);
839 break;
840 case S_IFDIR:
841 err = -EPERM;
842 break;
843 default:
844 err = -EINVAL;
845 }
846
847 if (!err)
848 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
849
850 return err;
851 }
852
853 /*
854 * Plain create. Intent create is handled in atomic_open.
855 */
856 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
857 umode_t mode, bool want_excl)
858 {
859 int rc;
860
861 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),"
862 "flags=%u, excl=%d\n",
863 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
864 dir->i_generation, dir, mode, want_excl);
865
866 rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
867
868 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
869
870 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
871 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
872
873 return rc;
874 }
875
876 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
877 const char *tgt, struct dentry *dchild)
878 {
879 int err;
880
881 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
882 name->len, name->name, dir->i_ino, dir->i_generation,
883 dir, 3000, tgt);
884
885 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
886 0, dchild, LUSTRE_OPC_SYMLINK);
887
888 if (!err)
889 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
890
891 return err;
892 }
893
894 static int ll_link_generic(struct inode *src, struct inode *dir,
895 struct qstr *name, struct dentry *dchild)
896 {
897 struct ll_sb_info *sbi = ll_i2sbi(dir);
898 struct ptlrpc_request *request = NULL;
899 struct md_op_data *op_data;
900 int err;
901
902 CDEBUG(D_VFSTRACE,
903 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
904 src->i_ino, src->i_generation, src, dir->i_ino,
905 dir->i_generation, dir, name->len, name->name);
906
907 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
908 0, LUSTRE_OPC_ANY, NULL);
909 if (IS_ERR(op_data))
910 return PTR_ERR(op_data);
911
912 err = md_link(sbi->ll_md_exp, op_data, &request);
913 ll_finish_md_op_data(op_data);
914 if (err)
915 GOTO(out, err);
916
917 ll_update_times(request, dir);
918 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
919 out:
920 ptlrpc_req_finished(request);
921 return err;
922 }
923
924 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
925 int mode, struct dentry *dchild)
926
927 {
928 int err;
929
930 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
931 name->len, name->name, dir->i_ino, dir->i_generation, dir);
932
933 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
934 mode &= ~current_umask();
935 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
936 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
937
938 if (!err)
939 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
940
941 return err;
942 }
943
944 /* Try to find the child dentry by its name.
945 If found, put the result fid into @fid. */
946 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
947 struct lu_fid *fid)
948 {
949 struct dentry *parent, *child;
950
951 parent = ll_d_hlist_entry(dir->i_dentry, struct dentry, d_alias);
952 child = d_lookup(parent, name);
953 if (child) {
954 if (child->d_inode)
955 *fid = *ll_inode2fid(child->d_inode);
956 dput(child);
957 }
958 }
959
960 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
961 struct dentry *dchild, struct qstr *name)
962 {
963 struct ptlrpc_request *request = NULL;
964 struct md_op_data *op_data;
965 int rc;
966
967 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
968 name->len, name->name, dir->i_ino, dir->i_generation, dir);
969
970 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
971 return -EBUSY;
972
973 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
974 S_IFDIR, LUSTRE_OPC_ANY, NULL);
975 if (IS_ERR(op_data))
976 return PTR_ERR(op_data);
977
978 ll_get_child_fid(dir, name, &op_data->op_fid3);
979 op_data->op_fid2 = op_data->op_fid3;
980 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
981 ll_finish_md_op_data(op_data);
982 if (rc == 0) {
983 ll_update_times(request, dir);
984 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
985 }
986
987 ptlrpc_req_finished(request);
988 return rc;
989 }
990
991 /**
992 * Remove dir entry
993 **/
994 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
995 {
996 struct ptlrpc_request *request = NULL;
997 struct md_op_data *op_data;
998 int rc;
999
1000 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1001 namelen, name, dir->i_ino, dir->i_generation, dir);
1002
1003 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1004 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1005 if (IS_ERR(op_data))
1006 return PTR_ERR(op_data);
1007 op_data->op_cli_flags |= CLI_RM_ENTRY;
1008 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1009 ll_finish_md_op_data(op_data);
1010 if (rc == 0) {
1011 ll_update_times(request, dir);
1012 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1013 }
1014
1015 ptlrpc_req_finished(request);
1016 return rc;
1017 }
1018
1019 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1020 {
1021 struct mdt_body *body;
1022 struct lov_mds_md *eadata;
1023 struct lov_stripe_md *lsm = NULL;
1024 struct obd_trans_info oti = { 0 };
1025 struct obdo *oa;
1026 struct obd_capa *oc = NULL;
1027 int rc;
1028
1029 /* req is swabbed so this is safe */
1030 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1031 if (!(body->valid & OBD_MD_FLEASIZE))
1032 return 0;
1033
1034 if (body->eadatasize == 0) {
1035 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1036 GOTO(out, rc = -EPROTO);
1037 }
1038
1039 /* The MDS sent back the EA because we unlinked the last reference
1040 * to this file. Use this EA to unlink the objects on the OST.
1041 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1042 * check it is complete and sensible. */
1043 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1044 body->eadatasize);
1045 LASSERT(eadata != NULL);
1046
1047 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1048 if (rc < 0) {
1049 CERROR("obd_unpackmd: %d\n", rc);
1050 GOTO(out, rc);
1051 }
1052 LASSERT(rc >= sizeof(*lsm));
1053
1054 OBDO_ALLOC(oa);
1055 if (oa == NULL)
1056 GOTO(out_free_memmd, rc = -ENOMEM);
1057
1058 oa->o_oi = lsm->lsm_oi;
1059 oa->o_mode = body->mode & S_IFMT;
1060 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1061
1062 if (body->valid & OBD_MD_FLCOOKIE) {
1063 oa->o_valid |= OBD_MD_FLCOOKIE;
1064 oti.oti_logcookies =
1065 req_capsule_server_sized_get(&request->rq_pill,
1066 &RMF_LOGCOOKIES,
1067 sizeof(struct llog_cookie) *
1068 lsm->lsm_stripe_count);
1069 if (oti.oti_logcookies == NULL) {
1070 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1071 body->valid &= ~OBD_MD_FLCOOKIE;
1072 }
1073 }
1074
1075 if (body->valid & OBD_MD_FLOSSCAPA) {
1076 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1077 if (rc)
1078 GOTO(out_free_memmd, rc);
1079 }
1080
1081 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1082 ll_i2mdexp(dir), oc);
1083 capa_put(oc);
1084 if (rc)
1085 CERROR("obd destroy objid "DOSTID" error %d\n",
1086 POSTID(&lsm->lsm_oi), rc);
1087 out_free_memmd:
1088 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1089 OBDO_FREE(oa);
1090 out:
1091 return rc;
1092 }
1093
1094 /* ll_unlink_generic() doesn't update the inode with the new link count.
1095 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1096 * is any lock existing. They will recycle dentries and inodes based upon locks
1097 * too. b=20433 */
1098 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1099 struct dentry *dchild, struct qstr *name)
1100 {
1101 struct ptlrpc_request *request = NULL;
1102 struct md_op_data *op_data;
1103 int rc;
1104 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1105 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1106
1107 /*
1108 * XXX: unlink bind mountpoint maybe call to here,
1109 * just check it as vfs_unlink does.
1110 */
1111 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1112 return -EBUSY;
1113
1114 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1115 name->len, 0, LUSTRE_OPC_ANY, NULL);
1116 if (IS_ERR(op_data))
1117 return PTR_ERR(op_data);
1118
1119 ll_get_child_fid(dir, name, &op_data->op_fid3);
1120 op_data->op_fid2 = op_data->op_fid3;
1121 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1122 ll_finish_md_op_data(op_data);
1123 if (rc)
1124 GOTO(out, rc);
1125
1126 ll_update_times(request, dir);
1127 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1128
1129 rc = ll_objects_destroy(request, dir);
1130 out:
1131 ptlrpc_req_finished(request);
1132 return rc;
1133 }
1134
1135 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1136 struct dentry *src_dchild, struct qstr *src_name,
1137 struct inode *tgt, struct dentry *tgt_dparent,
1138 struct dentry *tgt_dchild, struct qstr *tgt_name)
1139 {
1140 struct ptlrpc_request *request = NULL;
1141 struct ll_sb_info *sbi = ll_i2sbi(src);
1142 struct md_op_data *op_data;
1143 int err;
1144
1145 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1146 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1147 src->i_ino, src->i_generation, src, tgt_name->len,
1148 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1149
1150 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1151 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1152 return -EBUSY;
1153
1154 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1155 LUSTRE_OPC_ANY, NULL);
1156 if (IS_ERR(op_data))
1157 return PTR_ERR(op_data);
1158
1159 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1160 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1161 err = md_rename(sbi->ll_md_exp, op_data,
1162 src_name->name, src_name->len,
1163 tgt_name->name, tgt_name->len, &request);
1164 ll_finish_md_op_data(op_data);
1165 if (!err) {
1166 ll_update_times(request, src);
1167 ll_update_times(request, tgt);
1168 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1169 err = ll_objects_destroy(request, src);
1170 }
1171
1172 ptlrpc_req_finished(request);
1173
1174 return err;
1175 }
1176
1177 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1178 dev_t rdev)
1179 {
1180 return ll_mknod_generic(dir, &dchild->d_name, mode,
1181 old_encode_dev(rdev), dchild);
1182 }
1183
1184 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1185 {
1186 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1187 }
1188
1189 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1190 {
1191 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1192 }
1193
1194 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1195 {
1196 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1197 }
1198
1199 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1200 const char *oldname)
1201 {
1202 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1203 }
1204
1205 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1206 struct dentry *new_dentry)
1207 {
1208 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1209 new_dentry);
1210 }
1211
1212 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1213 struct inode *new_dir, struct dentry *new_dentry)
1214 {
1215 int err;
1216 err = ll_rename_generic(old_dir, NULL,
1217 old_dentry, &old_dentry->d_name,
1218 new_dir, NULL, new_dentry,
1219 &new_dentry->d_name);
1220 if (!err) {
1221 d_move(old_dentry, new_dentry);
1222 }
1223 return err;
1224 }
1225
1226 struct inode_operations ll_dir_inode_operations = {
1227 .mknod = ll_mknod,
1228 .atomic_open = ll_atomic_open,
1229 .lookup = ll_lookup_nd,
1230 .create = ll_create_nd,
1231 /* We need all these non-raw things for NFSD, to not patch it. */
1232 .unlink = ll_unlink,
1233 .mkdir = ll_mkdir,
1234 .rmdir = ll_rmdir,
1235 .symlink = ll_symlink,
1236 .link = ll_link,
1237 .rename = ll_rename,
1238 .setattr = ll_setattr,
1239 .getattr = ll_getattr,
1240 .permission = ll_inode_permission,
1241 .setxattr = ll_setxattr,
1242 .getxattr = ll_getxattr,
1243 .listxattr = ll_listxattr,
1244 .removexattr = ll_removexattr,
1245 .get_acl = ll_get_acl,
1246 };
1247
1248 struct inode_operations ll_special_inode_operations = {
1249 .setattr = ll_setattr,
1250 .getattr = ll_getattr,
1251 .permission = ll_inode_permission,
1252 .setxattr = ll_setxattr,
1253 .getxattr = ll_getxattr,
1254 .listxattr = ll_listxattr,
1255 .removexattr = ll_removexattr,
1256 .get_acl = ll_get_acl,
1257 };
Linux kernel - Deliverables
This page took 0.086811 seconds and 6 git commands to generate.