projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'iio-for-3.15c' of git://git.kernel.org/pub/scm/linux/kernel/git/jic23...
[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 struct lustre_handle lockh;
199 int rc;
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: rc = %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 __u64 bits = lock->l_policy_data.l_inodebits.bits;
213
214 /* Inode is set to lock->l_resource->lr_lvb_inode
215 * for mdc - bug 24555 */
216 LASSERT(lock->l_ast_data == NULL);
217
218 if (inode == NULL)
219 break;
220
221 /* Invalidate all dentries associated with this inode */
222 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
223
224 if (!fid_res_name_eq(ll_inode2fid(inode),
225 &lock->l_resource->lr_name)) {
226 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
227 PFID(ll_inode2fid(inode)), inode);
228 LBUG();
229 }
230
231 if (bits & MDS_INODELOCK_XATTR) {
232 ll_xattr_cache_destroy(inode);
233 bits &= ~MDS_INODELOCK_XATTR;
234 }
235
236 /* For OPEN locks we differentiate between lock modes
237 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
238 if (bits & MDS_INODELOCK_OPEN)
239 ll_have_md_lock(inode, &bits, lock->l_req_mode);
240
241 if (bits & MDS_INODELOCK_OPEN) {
242 fmode_t fmode;
243
244 switch (lock->l_req_mode) {
245 case LCK_CW:
246 fmode = FMODE_WRITE;
247 break;
248 case LCK_PR:
249 fmode = FMODE_EXEC;
250 break;
251 case LCK_CR:
252 fmode = FMODE_READ;
253 break;
254 default:
255 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
256 LBUG();
257 }
258
259 ll_md_real_close(inode, fmode);
260 }
261
262 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
263 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
264 ll_have_md_lock(inode, &bits, LCK_MINMODE);
265
266 if (bits & MDS_INODELOCK_LAYOUT) {
267 struct cl_object_conf conf = {
268 .coc_opc = OBJECT_CONF_INVALIDATE,
269 .coc_inode = inode,
270 };
271
272 rc = ll_layout_conf(inode, &conf);
273 if (rc < 0)
274 CDEBUG(D_INODE, "cannot invalidate layout of "
275 DFID": rc = %d\n",
276 PFID(ll_inode2fid(inode)), rc);
277 }
278
279 if (bits & MDS_INODELOCK_UPDATE) {
280 struct ll_inode_info *lli = ll_i2info(inode);
281
282 spin_lock(&lli->lli_lock);
283 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
284 spin_unlock(&lli->lli_lock);
285 }
286
287 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
288 CDEBUG(D_INODE, "invalidating inode %lu\n",
289 inode->i_ino);
290 truncate_inode_pages(inode->i_mapping, 0);
291 ll_invalidate_negative_children(inode);
292 }
293
294 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
295 inode->i_sb->s_root != NULL &&
296 inode != inode->i_sb->s_root->d_inode)
297 ll_invalidate_aliases(inode);
298
299 iput(inode);
300 break;
301 }
302 default:
303 LBUG();
304 }
305
306 return 0;
307 }
308
309 __u32 ll_i2suppgid(struct inode *i)
310 {
311 if (in_group_p(i->i_gid))
312 return (__u32)from_kgid(&init_user_ns, i->i_gid);
313 else
314 return (__u32)(-1);
315 }
316
317 /* Pack the required supplementary groups into the supplied groups array.
318 * If we don't need to use the groups from the target inode(s) then we
319 * instead pack one or more groups from the user's supplementary group
320 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
321 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
322 {
323 #if 0
324 int i;
325 #endif
326
327 LASSERT(i1 != NULL);
328 LASSERT(suppgids != NULL);
329
330 suppgids[0] = ll_i2suppgid(i1);
331
332 if (i2)
333 suppgids[1] = ll_i2suppgid(i2);
334 else
335 suppgids[1] = -1;
336
337 #if 0
338 for (i = 0; i < current_ngroups; i++) {
339 if (suppgids[0] == -1) {
340 if (current_groups[i] != suppgids[1])
341 suppgids[0] = current_groups[i];
342 continue;
343 }
344 if (suppgids[1] == -1) {
345 if (current_groups[i] != suppgids[0])
346 suppgids[1] = current_groups[i];
347 continue;
348 }
349 break;
350 }
351 #endif
352 }
353
354 /*
355 * try to reuse three types of dentry:
356 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
357 * by concurrent .revalidate).
358 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
359 * be cleared by others calling d_lustre_revalidate).
360 * 3. DISCONNECTED alias.
361 */
362 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
363 {
364 struct dentry *alias, *discon_alias, *invalid_alias;
365 struct ll_d_hlist_node *p;
366
367 if (ll_d_hlist_empty(&inode->i_dentry))
368 return NULL;
369
370 discon_alias = invalid_alias = NULL;
371
372 ll_lock_dcache(inode);
373 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
374 LASSERT(alias != dentry);
375
376 spin_lock(&alias->d_lock);
377 if (alias->d_flags & DCACHE_DISCONNECTED)
378 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
379 discon_alias = alias;
380 else if (alias->d_parent == dentry->d_parent &&
381 alias->d_name.hash == dentry->d_name.hash &&
382 alias->d_name.len == dentry->d_name.len &&
383 memcmp(alias->d_name.name, dentry->d_name.name,
384 dentry->d_name.len) == 0)
385 invalid_alias = alias;
386 spin_unlock(&alias->d_lock);
387
388 if (invalid_alias)
389 break;
390 }
391 alias = invalid_alias ?: discon_alias ?: NULL;
392 if (alias) {
393 spin_lock(&alias->d_lock);
394 dget_dlock(alias);
395 spin_unlock(&alias->d_lock);
396 }
397 ll_unlock_dcache(inode);
398
399 return alias;
400 }
401
402 /*
403 * Similar to d_splice_alias(), but lustre treats invalid alias
404 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
405 */
406 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
407 {
408 struct dentry *new;
409 int rc;
410
411 if (inode) {
412 new = ll_find_alias(inode, de);
413 if (new) {
414 rc = ll_d_init(new);
415 if (rc < 0) {
416 dput(new);
417 return ERR_PTR(rc);
418 }
419 d_move(new, de);
420 iput(inode);
421 CDEBUG(D_DENTRY,
422 "Reuse dentry %p inode %p refc %d flags %#x\n",
423 new, new->d_inode, d_count(new), new->d_flags);
424 return new;
425 }
426 }
427 rc = ll_d_init(de);
428 if (rc < 0)
429 return ERR_PTR(rc);
430 d_add(de, inode);
431 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
432 de, de->d_inode, d_count(de), de->d_flags);
433 return de;
434 }
435
436 int ll_lookup_it_finish(struct ptlrpc_request *request,
437 struct lookup_intent *it, void *data)
438 {
439 struct it_cb_data *icbd = data;
440 struct dentry **de = icbd->icbd_childp;
441 struct inode *parent = icbd->icbd_parent;
442 struct inode *inode = NULL;
443 __u64 bits = 0;
444 int rc;
445
446 /* NB 1 request reference will be taken away by ll_intent_lock()
447 * when I return */
448 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
449 it->d.lustre.it_disposition);
450 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
451 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
452 if (rc)
453 return rc;
454
455 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
456
457 /* We used to query real size from OSTs here, but actually
458 this is not needed. For stat() calls size would be updated
459 from subsequent do_revalidate()->ll_inode_revalidate_it() in
460 2.4 and
461 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
462 Everybody else who needs correct file size would call
463 ll_glimpse_size or some equivalent themselves anyway.
464 Also see bug 7198. */
465 }
466
467 /* Only hash *de if it is unhashed (new dentry).
468 * Atoimc_open may passin hashed dentries for open.
469 */
470 if (d_unhashed(*de)) {
471 struct dentry *alias;
472
473 alias = ll_splice_alias(inode, *de);
474 if (IS_ERR(alias))
475 return PTR_ERR(alias);
476 *de = alias;
477 } else if (!it_disposition(it, DISP_LOOKUP_NEG) &&
478 !it_disposition(it, DISP_OPEN_CREATE)) {
479 /* With DISP_OPEN_CREATE dentry will
480 instantiated in ll_create_it. */
481 LASSERT((*de)->d_inode == NULL);
482 d_instantiate(*de, inode);
483 }
484
485 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
486 /* we have lookup look - unhide dentry */
487 if (bits & MDS_INODELOCK_LOOKUP)
488 d_lustre_revalidate(*de);
489 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
490 /* If file created on server, don't depend on parent UPDATE
491 * lock to unhide it. It is left hidden and next lookup can
492 * find it in ll_splice_alias.
493 */
494 /* Check that parent has UPDATE lock. */
495 struct lookup_intent parent_it = {
496 .it_op = IT_GETATTR,
497 .d.lustre.it_lock_handle = 0 };
498
499 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
500 &ll_i2info(parent)->lli_fid, NULL)) {
501 d_lustre_revalidate(*de);
502 ll_intent_release(&parent_it);
503 }
504 }
505
506 return 0;
507 }
508
509 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
510 struct lookup_intent *it, int lookup_flags)
511 {
512 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
513 struct dentry *save = dentry, *retval;
514 struct ptlrpc_request *req = NULL;
515 struct md_op_data *op_data;
516 struct it_cb_data icbd;
517 __u32 opc;
518 int rc;
519
520 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
521 return ERR_PTR(-ENAMETOOLONG);
522
523 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
524 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
525 parent->i_generation, parent, LL_IT2STR(it));
526
527 if (d_mountpoint(dentry))
528 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
529
530 ll_frob_intent(&it, &lookup_it);
531
532 if (it->it_op == IT_GETATTR) {
533 rc = ll_statahead_enter(parent, &dentry, 0);
534 if (rc == 1) {
535 if (dentry == save)
536 GOTO(out, retval = NULL);
537 GOTO(out, retval = dentry);
538 }
539 }
540
541 icbd.icbd_childp = &dentry;
542 icbd.icbd_parent = parent;
543
544 if (it->it_op & IT_CREAT)
545 opc = LUSTRE_OPC_CREATE;
546 else
547 opc = LUSTRE_OPC_ANY;
548
549 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
550 dentry->d_name.len, lookup_flags, opc,
551 NULL);
552 if (IS_ERR(op_data))
553 return (void *)op_data;
554
555 /* enforce umask if acl disabled or MDS doesn't support umask */
556 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
557 it->it_create_mode &= ~current_umask();
558
559 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
560 lookup_flags, &req, ll_md_blocking_ast, 0);
561 ll_finish_md_op_data(op_data);
562 if (rc < 0)
563 GOTO(out, retval = ERR_PTR(rc));
564
565 rc = ll_lookup_it_finish(req, it, &icbd);
566 if (rc != 0) {
567 ll_intent_release(it);
568 GOTO(out, retval = ERR_PTR(rc));
569 }
570
571 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
572 !S_ISREG(dentry->d_inode->i_mode) &&
573 !S_ISDIR(dentry->d_inode->i_mode)) {
574 ll_release_openhandle(dentry, it);
575 }
576 ll_lookup_finish_locks(it, dentry);
577
578 if (dentry == save)
579 GOTO(out, retval = NULL);
580 else
581 GOTO(out, retval = dentry);
582 out:
583 if (req)
584 ptlrpc_req_finished(req);
585 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
586 ll_statahead_mark(parent, dentry);
587 return retval;
588 }
589
590 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
591 unsigned int flags)
592 {
593 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
594 struct dentry *de;
595
596 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),flags=%u\n",
597 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
598 parent->i_generation, parent, flags);
599
600 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
601 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
602 return NULL;
603
604 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
605 itp = NULL;
606 else
607 itp = &it;
608 de = ll_lookup_it(parent, dentry, itp, 0);
609
610 if (itp != NULL)
611 ll_intent_release(itp);
612
613 return de;
614 }
615
616 /*
617 * For cached negative dentry and new dentry, handle lookup/create/open
618 * together.
619 */
620 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
621 struct file *file, unsigned open_flags,
622 umode_t mode, int *opened)
623 {
624 struct lookup_intent *it;
625 struct dentry *de;
626 long long lookup_flags = LOOKUP_OPEN;
627 int rc = 0;
628
629 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),file %p,"
630 "open_flags %x,mode %x opened %d\n",
631 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
632 dir->i_generation, dir, file, open_flags, mode, *opened);
633
634 OBD_ALLOC(it, sizeof(*it));
635 if (!it)
636 return -ENOMEM;
637
638 it->it_op = IT_OPEN;
639 if (open_flags & O_CREAT) {
640 it->it_op |= IT_CREAT;
641 lookup_flags |= LOOKUP_CREATE;
642 }
643 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
644 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
645
646 /* Dentry added to dcache tree in ll_lookup_it */
647 de = ll_lookup_it(dir, dentry, it, lookup_flags);
648 if (IS_ERR(de))
649 rc = PTR_ERR(de);
650 else if (de != NULL)
651 dentry = de;
652
653 if (!rc) {
654 if (it_disposition(it, DISP_OPEN_CREATE)) {
655 /* Dentry instantiated in ll_create_it. */
656 rc = ll_create_it(dir, dentry, mode, it);
657 if (rc) {
658 /* We dget in ll_splice_alias. */
659 if (de != NULL)
660 dput(de);
661 goto out_release;
662 }
663
664 *opened |= FILE_CREATED;
665 }
666 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
667 /* Open dentry. */
668 if (S_ISFIFO(dentry->d_inode->i_mode)) {
669 /* We cannot call open here as it would
670 * deadlock.
671 */
672 if (it_disposition(it, DISP_ENQ_OPEN_REF))
673 ptlrpc_req_finished(
674 (struct ptlrpc_request *)
675 it->d.lustre.it_data);
676 rc = finish_no_open(file, de);
677 } else {
678 file->private_data = it;
679 rc = finish_open(file, dentry, NULL, opened);
680 /* We dget in ll_splice_alias. finish_open takes
681 * care of dget for fd open.
682 */
683 if (de != NULL)
684 dput(de);
685 }
686 } else {
687 rc = finish_no_open(file, de);
688 }
689 }
690
691 out_release:
692 ll_intent_release(it);
693 OBD_FREE(it, sizeof(*it));
694
695 return rc;
696 }
697
698
699 /* We depend on "mode" being set with the proper file type/umask by now */
700 static struct inode *ll_create_node(struct inode *dir, const char *name,
701 int namelen, const void *data, int datalen,
702 int mode, __u64 extra,
703 struct lookup_intent *it)
704 {
705 struct inode *inode = NULL;
706 struct ptlrpc_request *request = NULL;
707 struct ll_sb_info *sbi = ll_i2sbi(dir);
708 int rc;
709
710 LASSERT(it && it->d.lustre.it_disposition);
711
712 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
713 request = it->d.lustre.it_data;
714 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
715 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
716 if (rc)
717 GOTO(out, inode = ERR_PTR(rc));
718
719 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
720
721 /* We asked for a lock on the directory, but were granted a
722 * lock on the inode. Since we finally have an inode pointer,
723 * stuff it in the lock. */
724 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
725 inode, inode->i_ino, inode->i_generation);
726 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
727 out:
728 ptlrpc_req_finished(request);
729 return inode;
730 }
731
732 /*
733 * By the time this is called, we already have created the directory cache
734 * entry for the new file, but it is so far negative - it has no inode.
735 *
736 * We defer creating the OBD object(s) until open, to keep the intent and
737 * non-intent code paths similar, and also because we do not have the MDS
738 * inode number before calling ll_create_node() (which is needed for LOV),
739 * so we would need to do yet another RPC to the MDS to store the LOV EA
740 * data on the MDS. If needed, we would pass the PACKED lmm as data and
741 * lmm_size in datalen (the MDS still has code which will handle that).
742 *
743 * If the create succeeds, we fill in the inode information
744 * with d_instantiate().
745 */
746 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
747 struct lookup_intent *it)
748 {
749 struct inode *inode;
750 int rc = 0;
751
752 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
753 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
754 dir->i_generation, dir, LL_IT2STR(it));
755
756 rc = it_open_error(DISP_OPEN_CREATE, it);
757 if (rc)
758 return rc;
759
760 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
761 NULL, 0, mode, 0, it);
762 if (IS_ERR(inode))
763 return PTR_ERR(inode);
764
765 if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
766 ll_i2info(inode)->lli_volatile = true;
767
768 d_instantiate(dentry, inode);
769 return 0;
770 }
771
772 static void ll_update_times(struct ptlrpc_request *request,
773 struct inode *inode)
774 {
775 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
776 &RMF_MDT_BODY);
777
778 LASSERT(body);
779 if (body->valid & OBD_MD_FLMTIME &&
780 body->mtime > LTIME_S(inode->i_mtime)) {
781 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
782 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
783 LTIME_S(inode->i_mtime) = body->mtime;
784 }
785 if (body->valid & OBD_MD_FLCTIME &&
786 body->ctime > LTIME_S(inode->i_ctime))
787 LTIME_S(inode->i_ctime) = body->ctime;
788 }
789
790 static int ll_new_node(struct inode *dir, struct qstr *name,
791 const char *tgt, int mode, int rdev,
792 struct dentry *dchild, __u32 opc)
793 {
794 struct ptlrpc_request *request = NULL;
795 struct md_op_data *op_data;
796 struct inode *inode = NULL;
797 struct ll_sb_info *sbi = ll_i2sbi(dir);
798 int tgt_len = 0;
799 int err;
800
801 if (unlikely(tgt != NULL))
802 tgt_len = strlen(tgt) + 1;
803
804 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
805 name->len, 0, opc, NULL);
806 if (IS_ERR(op_data))
807 GOTO(err_exit, err = PTR_ERR(op_data));
808
809 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
810 from_kuid(&init_user_ns, current_fsuid()),
811 from_kgid(&init_user_ns, current_fsgid()),
812 cfs_curproc_cap_pack(), rdev, &request);
813 ll_finish_md_op_data(op_data);
814 if (err)
815 GOTO(err_exit, err);
816
817 ll_update_times(request, dir);
818
819 if (dchild) {
820 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
821 if (err)
822 GOTO(err_exit, err);
823
824 d_instantiate(dchild, inode);
825 }
826 err_exit:
827 ptlrpc_req_finished(request);
828
829 return err;
830 }
831
832 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
833 unsigned rdev, struct dentry *dchild)
834 {
835 int err;
836
837 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
838 name->len, name->name, dir->i_ino, dir->i_generation, dir,
839 mode, rdev);
840
841 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
842 mode &= ~current_umask();
843
844 switch (mode & S_IFMT) {
845 case 0:
846 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
847 case S_IFREG:
848 case S_IFCHR:
849 case S_IFBLK:
850 case S_IFIFO:
851 case S_IFSOCK:
852 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
853 LUSTRE_OPC_MKNOD);
854 break;
855 case S_IFDIR:
856 err = -EPERM;
857 break;
858 default:
859 err = -EINVAL;
860 }
861
862 if (!err)
863 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
864
865 return err;
866 }
867
868 /*
869 * Plain create. Intent create is handled in atomic_open.
870 */
871 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
872 umode_t mode, bool want_excl)
873 {
874 int rc;
875
876 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),"
877 "flags=%u, excl=%d\n",
878 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
879 dir->i_generation, dir, mode, want_excl);
880
881 rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
882
883 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
884
885 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
886 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
887
888 return rc;
889 }
890
891 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
892 const char *tgt, struct dentry *dchild)
893 {
894 int err;
895
896 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
897 name->len, name->name, dir->i_ino, dir->i_generation,
898 dir, 3000, tgt);
899
900 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
901 0, dchild, LUSTRE_OPC_SYMLINK);
902
903 if (!err)
904 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
905
906 return err;
907 }
908
909 static int ll_link_generic(struct inode *src, struct inode *dir,
910 struct qstr *name, struct dentry *dchild)
911 {
912 struct ll_sb_info *sbi = ll_i2sbi(dir);
913 struct ptlrpc_request *request = NULL;
914 struct md_op_data *op_data;
915 int err;
916
917 CDEBUG(D_VFSTRACE,
918 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
919 src->i_ino, src->i_generation, src, dir->i_ino,
920 dir->i_generation, dir, name->len, name->name);
921
922 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
923 0, LUSTRE_OPC_ANY, NULL);
924 if (IS_ERR(op_data))
925 return PTR_ERR(op_data);
926
927 err = md_link(sbi->ll_md_exp, op_data, &request);
928 ll_finish_md_op_data(op_data);
929 if (err)
930 GOTO(out, err);
931
932 ll_update_times(request, dir);
933 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
934 out:
935 ptlrpc_req_finished(request);
936 return err;
937 }
938
939 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
940 int mode, struct dentry *dchild)
941
942 {
943 int err;
944
945 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
946 name->len, name->name, dir->i_ino, dir->i_generation, dir);
947
948 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
949 mode &= ~current_umask();
950 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
951 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
952
953 if (!err)
954 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
955
956 return err;
957 }
958
959 /* Try to find the child dentry by its name.
960 If found, put the result fid into @fid. */
961 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
962 struct lu_fid *fid)
963 {
964 struct dentry *parent, *child;
965
966 parent = ll_d_hlist_entry(dir->i_dentry, struct dentry, d_alias);
967 child = d_lookup(parent, name);
968 if (child) {
969 if (child->d_inode)
970 *fid = *ll_inode2fid(child->d_inode);
971 dput(child);
972 }
973 }
974
975 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
976 struct dentry *dchild, struct qstr *name)
977 {
978 struct ptlrpc_request *request = NULL;
979 struct md_op_data *op_data;
980 int rc;
981
982 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
983 name->len, name->name, dir->i_ino, dir->i_generation, dir);
984
985 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
986 return -EBUSY;
987
988 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
989 S_IFDIR, LUSTRE_OPC_ANY, NULL);
990 if (IS_ERR(op_data))
991 return PTR_ERR(op_data);
992
993 ll_get_child_fid(dir, name, &op_data->op_fid3);
994 op_data->op_fid2 = op_data->op_fid3;
995 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
996 ll_finish_md_op_data(op_data);
997 if (rc == 0) {
998 ll_update_times(request, dir);
999 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1000 }
1001
1002 ptlrpc_req_finished(request);
1003 return rc;
1004 }
1005
1006 /**
1007 * Remove dir entry
1008 **/
1009 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1010 {
1011 struct ptlrpc_request *request = NULL;
1012 struct md_op_data *op_data;
1013 int rc;
1014
1015 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1016 namelen, name, dir->i_ino, dir->i_generation, dir);
1017
1018 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1019 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1020 if (IS_ERR(op_data))
1021 return PTR_ERR(op_data);
1022 op_data->op_cli_flags |= CLI_RM_ENTRY;
1023 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1024 ll_finish_md_op_data(op_data);
1025 if (rc == 0) {
1026 ll_update_times(request, dir);
1027 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1028 }
1029
1030 ptlrpc_req_finished(request);
1031 return rc;
1032 }
1033
1034 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1035 {
1036 struct mdt_body *body;
1037 struct lov_mds_md *eadata;
1038 struct lov_stripe_md *lsm = NULL;
1039 struct obd_trans_info oti = { 0 };
1040 struct obdo *oa;
1041 struct obd_capa *oc = NULL;
1042 int rc;
1043
1044 /* req is swabbed so this is safe */
1045 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1046 if (!(body->valid & OBD_MD_FLEASIZE))
1047 return 0;
1048
1049 if (body->eadatasize == 0) {
1050 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1051 GOTO(out, rc = -EPROTO);
1052 }
1053
1054 /* The MDS sent back the EA because we unlinked the last reference
1055 * to this file. Use this EA to unlink the objects on the OST.
1056 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1057 * check it is complete and sensible. */
1058 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1059 body->eadatasize);
1060 LASSERT(eadata != NULL);
1061
1062 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1063 if (rc < 0) {
1064 CERROR("obd_unpackmd: %d\n", rc);
1065 GOTO(out, rc);
1066 }
1067 LASSERT(rc >= sizeof(*lsm));
1068
1069 OBDO_ALLOC(oa);
1070 if (oa == NULL)
1071 GOTO(out_free_memmd, rc = -ENOMEM);
1072
1073 oa->o_oi = lsm->lsm_oi;
1074 oa->o_mode = body->mode & S_IFMT;
1075 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1076
1077 if (body->valid & OBD_MD_FLCOOKIE) {
1078 oa->o_valid |= OBD_MD_FLCOOKIE;
1079 oti.oti_logcookies =
1080 req_capsule_server_sized_get(&request->rq_pill,
1081 &RMF_LOGCOOKIES,
1082 sizeof(struct llog_cookie) *
1083 lsm->lsm_stripe_count);
1084 if (oti.oti_logcookies == NULL) {
1085 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1086 body->valid &= ~OBD_MD_FLCOOKIE;
1087 }
1088 }
1089
1090 if (body->valid & OBD_MD_FLOSSCAPA) {
1091 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1092 if (rc)
1093 GOTO(out_free_memmd, rc);
1094 }
1095
1096 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1097 ll_i2mdexp(dir), oc);
1098 capa_put(oc);
1099 if (rc)
1100 CERROR("obd destroy objid "DOSTID" error %d\n",
1101 POSTID(&lsm->lsm_oi), rc);
1102 out_free_memmd:
1103 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1104 OBDO_FREE(oa);
1105 out:
1106 return rc;
1107 }
1108
1109 /* ll_unlink_generic() doesn't update the inode with the new link count.
1110 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1111 * is any lock existing. They will recycle dentries and inodes based upon locks
1112 * too. b=20433 */
1113 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1114 struct dentry *dchild, struct qstr *name)
1115 {
1116 struct ptlrpc_request *request = NULL;
1117 struct md_op_data *op_data;
1118 int rc;
1119 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1120 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1121
1122 /*
1123 * XXX: unlink bind mountpoint maybe call to here,
1124 * just check it as vfs_unlink does.
1125 */
1126 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1127 return -EBUSY;
1128
1129 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1130 name->len, 0, LUSTRE_OPC_ANY, NULL);
1131 if (IS_ERR(op_data))
1132 return PTR_ERR(op_data);
1133
1134 ll_get_child_fid(dir, name, &op_data->op_fid3);
1135 op_data->op_fid2 = op_data->op_fid3;
1136 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1137 ll_finish_md_op_data(op_data);
1138 if (rc)
1139 GOTO(out, rc);
1140
1141 ll_update_times(request, dir);
1142 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1143
1144 rc = ll_objects_destroy(request, dir);
1145 out:
1146 ptlrpc_req_finished(request);
1147 return rc;
1148 }
1149
1150 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1151 struct dentry *src_dchild, struct qstr *src_name,
1152 struct inode *tgt, struct dentry *tgt_dparent,
1153 struct dentry *tgt_dchild, struct qstr *tgt_name)
1154 {
1155 struct ptlrpc_request *request = NULL;
1156 struct ll_sb_info *sbi = ll_i2sbi(src);
1157 struct md_op_data *op_data;
1158 int err;
1159
1160 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1161 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1162 src->i_ino, src->i_generation, src, tgt_name->len,
1163 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1164
1165 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1166 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1167 return -EBUSY;
1168
1169 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1170 LUSTRE_OPC_ANY, NULL);
1171 if (IS_ERR(op_data))
1172 return PTR_ERR(op_data);
1173
1174 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1175 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1176 err = md_rename(sbi->ll_md_exp, op_data,
1177 src_name->name, src_name->len,
1178 tgt_name->name, tgt_name->len, &request);
1179 ll_finish_md_op_data(op_data);
1180 if (!err) {
1181 ll_update_times(request, src);
1182 ll_update_times(request, tgt);
1183 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1184 err = ll_objects_destroy(request, src);
1185 }
1186
1187 ptlrpc_req_finished(request);
1188
1189 return err;
1190 }
1191
1192 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1193 dev_t rdev)
1194 {
1195 return ll_mknod_generic(dir, &dchild->d_name, mode,
1196 old_encode_dev(rdev), dchild);
1197 }
1198
1199 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1200 {
1201 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1202 }
1203
1204 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1205 {
1206 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1207 }
1208
1209 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1210 {
1211 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1212 }
1213
1214 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1215 const char *oldname)
1216 {
1217 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1218 }
1219
1220 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1221 struct dentry *new_dentry)
1222 {
1223 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1224 new_dentry);
1225 }
1226
1227 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1228 struct inode *new_dir, struct dentry *new_dentry)
1229 {
1230 int err;
1231 err = ll_rename_generic(old_dir, NULL,
1232 old_dentry, &old_dentry->d_name,
1233 new_dir, NULL, new_dentry,
1234 &new_dentry->d_name);
1235 if (!err) {
1236 d_move(old_dentry, new_dentry);
1237 }
1238 return err;
1239 }
1240
1241 struct inode_operations ll_dir_inode_operations = {
1242 .mknod = ll_mknod,
1243 .atomic_open = ll_atomic_open,
1244 .lookup = ll_lookup_nd,
1245 .create = ll_create_nd,
1246 /* We need all these non-raw things for NFSD, to not patch it. */
1247 .unlink = ll_unlink,
1248 .mkdir = ll_mkdir,
1249 .rmdir = ll_rmdir,
1250 .symlink = ll_symlink,
1251 .link = ll_link,
1252 .rename = ll_rename,
1253 .setattr = ll_setattr,
1254 .getattr = ll_getattr,
1255 .permission = ll_inode_permission,
1256 .setxattr = ll_setxattr,
1257 .getxattr = ll_getxattr,
1258 .listxattr = ll_listxattr,
1259 .removexattr = ll_removexattr,
1260 .get_acl = ll_get_acl,
1261 };
1262
1263 struct inode_operations ll_special_inode_operations = {
1264 .setattr = ll_setattr,
1265 .getattr = ll_getattr,
1266 .permission = ll_inode_permission,
1267 .setxattr = ll_setxattr,
1268 .getxattr = ll_getxattr,
1269 .listxattr = ll_listxattr,
1270 .removexattr = ll_removexattr,
1271 .get_acl = ll_get_acl,
1272 };
Linux kernel - Deliverables
This page took 0.059896 seconds and 6 git commands to generate.