4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 #include <linux/sched.h>
40 #include <linux/quotaops.h>
41 #include <linux/highmem.h>
42 #include <linux/pagemap.h>
43 #include <linux/security.h>
45 #define DEBUG_SUBSYSTEM S_LLITE
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"
54 static int ll_create_it(struct inode *, struct dentry *,
55 int, struct lookup_intent *);
58 * Check if we have something mounted at the named dchild.
59 * In such a case there would always be dentry present.
61 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
66 if (unlikely(dchild)) {
67 mounted = d_mountpoint(dchild);
69 dchild = d_lookup(dparent, name);
71 mounted = d_mountpoint(dchild);
78 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
82 ldlm_lock_decref(lockh, mode);
88 /* called from iget5_locked->find_inode() under inode_lock spinlock */
89 static int ll_test_inode(struct inode *inode, void *opaque)
91 struct ll_inode_info *lli = ll_i2info(inode);
92 struct lustre_md *md = opaque;
94 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
95 CERROR("MDS body missing FID\n");
99 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
105 static int ll_set_inode(struct inode *inode, void *opaque)
107 struct ll_inode_info *lli = ll_i2info(inode);
108 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
110 if (unlikely(!(body->valid & OBD_MD_FLID))) {
111 CERROR("MDS body missing FID\n");
115 lli->lli_fid = body->fid1;
116 if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
117 CERROR("Can not initialize inode "DFID" without object type: "
118 "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
122 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
123 if (unlikely(inode->i_mode == 0)) {
124 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
135 * Get an inode by inode number (already instantiated by the intent lookup).
136 * Returns inode or NULL
138 struct inode *ll_iget(struct super_block *sb, ino_t hash,
139 struct lustre_md *md)
145 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
148 if (inode->i_state & I_NEW) {
151 ll_read_inode2(inode, md);
152 if (S_ISREG(inode->i_mode) &&
153 ll_i2info(inode)->lli_clob == NULL) {
155 "%s: apply lsm %p to inode "DFID".\n",
156 ll_get_fsname(sb, NULL, 0), md->lsm,
157 PFID(ll_inode2fid(inode)));
158 rc = cl_file_inode_init(inode, md);
161 make_bad_inode(inode);
162 unlock_new_inode(inode);
166 unlock_new_inode(inode);
167 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
168 ll_update_inode(inode, md);
169 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
170 inode, PFID(&md->body->fid1));
175 static void ll_invalidate_negative_children(struct inode *dir)
177 struct dentry *dentry, *tmp_subdir;
178 struct ll_d_hlist_node *p;
181 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
182 spin_lock(&dentry->d_lock);
183 if (!list_empty(&dentry->d_subdirs)) {
184 struct dentry *child;
186 list_for_each_entry_safe(child, tmp_subdir,
189 if (child->d_inode == NULL)
190 d_lustre_invalidate(child, 1);
193 spin_unlock(&dentry->d_lock);
195 ll_unlock_dcache(dir);
198 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
199 void *data, int flag)
202 struct lustre_handle lockh;
206 case LDLM_CB_BLOCKING:
207 ldlm_lock2handle(lock, &lockh);
208 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
210 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
214 case LDLM_CB_CANCELING: {
215 struct inode *inode = ll_inode_from_resource_lock(lock);
216 struct ll_inode_info *lli;
217 __u64 bits = lock->l_policy_data.l_inodebits.bits;
219 ldlm_mode_t mode = lock->l_req_mode;
221 /* Inode is set to lock->l_resource->lr_lvb_inode
222 * for mdc - bug 24555 */
223 LASSERT(lock->l_ast_data == NULL);
225 /* Invalidate all dentries associated with this inode */
229 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
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);
236 if (bits & MDS_INODELOCK_OPEN)
237 ll_have_md_lock(inode, &bits, mode);
239 fid = ll_inode2fid(inode);
240 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
241 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
242 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
243 LDLM_ERROR(lock, "data mismatch with object "
244 DFID" (%p)", PFID(fid), inode);
247 if (bits & MDS_INODELOCK_OPEN) {
249 switch (lock->l_req_mode) {
260 CERROR("Unexpected lock mode for OPEN lock "
261 "%d, inode %ld\n", lock->l_req_mode,
264 ll_md_real_close(inode, flags);
267 lli = ll_i2info(inode);
268 if (bits & MDS_INODELOCK_LAYOUT) {
269 struct cl_object_conf conf = { { 0 } };
271 conf.coc_opc = OBJECT_CONF_INVALIDATE;
272 conf.coc_inode = inode;
273 rc = ll_layout_conf(inode, &conf);
275 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
278 if (bits & MDS_INODELOCK_UPDATE)
279 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
281 if (S_ISDIR(inode->i_mode) &&
282 (bits & MDS_INODELOCK_UPDATE)) {
283 CDEBUG(D_INODE, "invalidating inode %lu\n",
285 truncate_inode_pages(inode->i_mapping, 0);
286 ll_invalidate_negative_children(inode);
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);
303 __u32 ll_i2suppgid(struct inode *i)
305 if (current_is_in_group(i->i_gid))
306 return (__u32)i->i_gid;
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)
322 LASSERT(suppgids != NULL);
324 suppgids[0] = ll_i2suppgid(i1);
327 suppgids[1] = ll_i2suppgid(i2);
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];
338 if (suppgids[1] == -1) {
339 if (current_groups[i] != suppgids[0])
340 suppgids[1] = current_groups[i];
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.
356 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
358 struct dentry *alias, *discon_alias, *invalid_alias;
359 struct ll_d_hlist_node *p;
361 if (ll_d_hlist_empty(&inode->i_dentry))
364 discon_alias = invalid_alias = NULL;
366 ll_lock_dcache(inode);
367 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
368 LASSERT(alias != dentry);
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);
385 alias = invalid_alias ?: discon_alias ?: NULL;
387 spin_lock(&alias->d_lock);
389 spin_unlock(&alias->d_lock);
391 ll_unlock_dcache(inode);
397 * Similar to d_splice_alias(), but lustre treats invalid alias
398 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
400 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
405 new = ll_find_alias(inode, de);
407 ll_dops_init(new, 1, 1);
411 "Reuse dentry %p inode %p refc %d flags %#x\n",
412 new, new->d_inode, d_count(new), new->d_flags);
416 ll_dops_init(de, 1, 1);
417 __d_lustre_invalidate(de);
419 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
420 de, de->d_inode, d_count(de), de->d_flags);
424 int ll_lookup_it_finish(struct ptlrpc_request *request,
425 struct lookup_intent *it, void *data)
427 struct it_cb_data *icbd = data;
428 struct dentry **de = icbd->icbd_childp;
429 struct inode *parent = icbd->icbd_parent;
430 struct inode *inode = NULL;
435 /* NB 1 request reference will be taken away by ll_intent_lock()
437 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
438 it->d.lustre.it_disposition);
439 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
440 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
444 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
446 /* We used to query real size from OSTs here, but actually
447 this is not needed. For stat() calls size would be updated
448 from subsequent do_revalidate()->ll_inode_revalidate_it() in
450 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
451 Everybody else who needs correct file size would call
452 ll_glimpse_size or some equivalent themselves anyway.
453 Also see bug 7198. */
456 /* Only hash *de if it is unhashed (new dentry).
457 * Atoimc_open may passin hashed dentries for open.
460 *de = ll_splice_alias(inode, *de);
462 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
463 /* we have lookup look - unhide dentry */
464 if (bits & MDS_INODELOCK_LOOKUP)
465 d_lustre_revalidate(*de);
466 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
467 /* If file created on server, don't depend on parent UPDATE
468 * lock to unhide it. It is left hidden and next lookup can
469 * find it in ll_splice_alias.
471 /* Check that parent has UPDATE lock. */
472 struct lookup_intent parent_it = {
474 .d.lustre.it_lock_handle = 0 };
476 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
477 &ll_i2info(parent)->lli_fid, NULL)) {
478 d_lustre_revalidate(*de);
479 ll_intent_release(&parent_it);
486 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
487 struct lookup_intent *it, int lookup_flags)
489 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
490 struct dentry *save = dentry, *retval;
491 struct ptlrpc_request *req = NULL;
492 struct md_op_data *op_data;
493 struct it_cb_data icbd;
498 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
499 RETURN(ERR_PTR(-ENAMETOOLONG));
501 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
502 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
503 parent->i_generation, parent, LL_IT2STR(it));
505 if (d_mountpoint(dentry))
506 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
508 ll_frob_intent(&it, &lookup_it);
510 /* As do_lookup is called before follow_mount, root dentry may be left
511 * not valid, revalidate it here. */
512 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
513 (it->it_op & (IT_OPEN | IT_CREAT))) {
514 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
515 MDS_INODELOCK_LOOKUP);
520 if (it->it_op == IT_GETATTR) {
521 rc = ll_statahead_enter(parent, &dentry, 0);
524 GOTO(out, retval = NULL);
525 GOTO(out, retval = dentry);
529 icbd.icbd_childp = &dentry;
530 icbd.icbd_parent = parent;
532 if (it->it_op & IT_CREAT ||
533 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
534 opc = LUSTRE_OPC_CREATE;
536 opc = LUSTRE_OPC_ANY;
538 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
539 dentry->d_name.len, lookup_flags, opc,
542 RETURN((void *)op_data);
544 /* enforce umask if acl disabled or MDS doesn't support umask */
545 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
546 it->it_create_mode &= ~current_umask();
548 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
549 lookup_flags, &req, ll_md_blocking_ast, 0);
550 ll_finish_md_op_data(op_data);
552 GOTO(out, retval = ERR_PTR(rc));
554 rc = ll_lookup_it_finish(req, it, &icbd);
556 ll_intent_release(it);
557 GOTO(out, retval = ERR_PTR(rc));
560 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
561 !S_ISREG(dentry->d_inode->i_mode) &&
562 !S_ISDIR(dentry->d_inode->i_mode)) {
563 ll_release_openhandle(dentry, it);
565 ll_lookup_finish_locks(it, dentry);
568 GOTO(out, retval = NULL);
570 GOTO(out, retval = dentry);
573 ptlrpc_req_finished(req);
574 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
575 ll_statahead_mark(parent, dentry);
579 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
582 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
585 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),flags=%u\n",
586 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
587 parent->i_generation, parent, flags);
589 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
590 if ((flags & LOOKUP_CREATE ) && !(flags & LOOKUP_OPEN)) {
591 ll_dops_init(dentry, 1, 1);
592 __d_lustre_invalidate(dentry);
597 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
601 de = ll_lookup_it(parent, dentry, itp, 0);
604 ll_intent_release(itp);
610 * For cached negative dentry and new dentry, handle lookup/create/open
613 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
614 struct file *file, unsigned open_flags,
615 umode_t mode, int *opened)
617 struct lookup_intent *it;
619 long long lookup_flags = LOOKUP_OPEN;
623 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),file %p,"
624 "open_flags %x,mode %x opened %d\n",
625 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
626 dir->i_generation, dir, file, open_flags, mode, *opened);
628 OBD_ALLOC(it, sizeof(*it));
634 it->it_op |= IT_CREAT;
635 lookup_flags |= LOOKUP_CREATE;
637 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
638 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
640 /* Dentry added to dcache tree in ll_lookup_it */
641 de = ll_lookup_it(dir, dentry, it, lookup_flags);
648 if (it_disposition(it, DISP_OPEN_CREATE)) {
649 /* Dentry instantiated in ll_create_it. */
650 rc = ll_create_it(dir, dentry, mode, it);
652 /* We dget in ll_splice_alias. */
658 *opened |= FILE_CREATED;
660 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
662 if (S_ISFIFO(dentry->d_inode->i_mode)) {
663 /* We cannot call open here as it would
666 if (it_disposition(it, DISP_ENQ_OPEN_REF))
668 (struct ptlrpc_request *)
669 it->d.lustre.it_data);
670 rc = finish_no_open(file, de);
672 file->private_data = it;
673 rc = finish_open(file, dentry, NULL, opened);
674 /* We dget in ll_splice_alias. finish_open takes
675 * care of dget for fd open.
681 rc = finish_no_open(file, de);
686 ll_intent_release(it);
687 OBD_FREE(it, sizeof(*it));
693 /* We depend on "mode" being set with the proper file type/umask by now */
694 static struct inode *ll_create_node(struct inode *dir, const char *name,
695 int namelen, const void *data, int datalen,
696 int mode, __u64 extra,
697 struct lookup_intent *it)
699 struct inode *inode = NULL;
700 struct ptlrpc_request *request = NULL;
701 struct ll_sb_info *sbi = ll_i2sbi(dir);
705 LASSERT(it && it->d.lustre.it_disposition);
707 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
708 request = it->d.lustre.it_data;
709 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
710 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
712 GOTO(out, inode = ERR_PTR(rc));
714 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
716 /* We asked for a lock on the directory, but were granted a
717 * lock on the inode. Since we finally have an inode pointer,
718 * stuff it in the lock. */
719 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
720 inode, inode->i_ino, inode->i_generation);
721 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
724 ptlrpc_req_finished(request);
729 * By the time this is called, we already have created the directory cache
730 * entry for the new file, but it is so far negative - it has no inode.
732 * We defer creating the OBD object(s) until open, to keep the intent and
733 * non-intent code paths similar, and also because we do not have the MDS
734 * inode number before calling ll_create_node() (which is needed for LOV),
735 * so we would need to do yet another RPC to the MDS to store the LOV EA
736 * data on the MDS. If needed, we would pass the PACKED lmm as data and
737 * lmm_size in datalen (the MDS still has code which will handle that).
739 * If the create succeeds, we fill in the inode information
740 * with d_instantiate().
742 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
743 struct lookup_intent *it)
749 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
750 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
751 dir->i_generation, dir, LL_IT2STR(it));
753 rc = it_open_error(DISP_OPEN_CREATE, it);
757 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
758 NULL, 0, mode, 0, it);
760 RETURN(PTR_ERR(inode));
762 if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
763 ll_i2info(inode)->lli_volatile = true;
765 d_instantiate(dentry, inode);
769 static void ll_update_times(struct ptlrpc_request *request,
772 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
776 if (body->valid & OBD_MD_FLMTIME &&
777 body->mtime > LTIME_S(inode->i_mtime)) {
778 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
779 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
780 LTIME_S(inode->i_mtime) = body->mtime;
782 if (body->valid & OBD_MD_FLCTIME &&
783 body->ctime > LTIME_S(inode->i_ctime))
784 LTIME_S(inode->i_ctime) = body->ctime;
787 static int ll_new_node(struct inode *dir, struct qstr *name,
788 const char *tgt, int mode, int rdev,
789 struct dentry *dchild, __u32 opc)
791 struct ptlrpc_request *request = NULL;
792 struct md_op_data *op_data;
793 struct inode *inode = NULL;
794 struct ll_sb_info *sbi = ll_i2sbi(dir);
799 if (unlikely(tgt != NULL))
800 tgt_len = strlen(tgt) + 1;
802 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
803 name->len, 0, opc, NULL);
805 GOTO(err_exit, err = PTR_ERR(op_data));
807 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
808 current_fsuid(), current_fsgid(),
809 cfs_curproc_cap_pack(), rdev, &request);
810 ll_finish_md_op_data(op_data);
814 ll_update_times(request, dir);
817 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
821 d_instantiate(dchild, inode);
825 ptlrpc_req_finished(request);
830 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
831 unsigned rdev, struct dentry *dchild)
836 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
837 name->len, name->name, dir->i_ino, dir->i_generation, dir,
840 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
841 mode &= ~current_umask();
843 switch (mode & S_IFMT) {
845 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
851 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
862 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
868 * Plain create. Intent create is handled in atomic_open.
870 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
871 umode_t mode, bool want_excl)
875 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),"
876 "flags=%u, excl=%d\n",
877 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
878 dir->i_generation, dir, mode, want_excl);
880 rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
882 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
884 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
885 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
890 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
891 const char *tgt, struct dentry *dchild)
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,
900 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
901 0, dchild, LUSTRE_OPC_SYMLINK);
904 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
909 static int ll_link_generic(struct inode *src, struct inode *dir,
910 struct qstr *name, struct dentry *dchild)
912 struct ll_sb_info *sbi = ll_i2sbi(dir);
913 struct ptlrpc_request *request = NULL;
914 struct md_op_data *op_data;
919 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
920 src->i_ino, src->i_generation, src, dir->i_ino,
921 dir->i_generation, dir, name->len, name->name);
923 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
924 0, LUSTRE_OPC_ANY, NULL);
926 RETURN(PTR_ERR(op_data));
928 err = md_link(sbi->ll_md_exp, op_data, &request);
929 ll_finish_md_op_data(op_data);
933 ll_update_times(request, dir);
934 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
937 ptlrpc_req_finished(request);
941 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
942 int mode, struct dentry *dchild)
948 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
949 name->len, name->name, dir->i_ino, dir->i_generation, dir);
951 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
952 mode &= ~current_umask();
953 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
954 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
957 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
962 /* Try to find the child dentry by its name.
963 If found, put the result fid into @fid. */
964 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
967 struct dentry *parent, *child;
969 parent = ll_d_hlist_entry(dir->i_dentry, struct dentry, d_alias);
970 child = d_lookup(parent, name);
973 *fid = *ll_inode2fid(child->d_inode);
978 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
979 struct dentry *dchild, struct qstr *name)
981 struct ptlrpc_request *request = NULL;
982 struct md_op_data *op_data;
986 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
987 name->len, name->name, dir->i_ino, dir->i_generation, dir);
989 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
992 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
993 S_IFDIR, LUSTRE_OPC_ANY, NULL);
995 RETURN(PTR_ERR(op_data));
997 ll_get_child_fid(dir, name, &op_data->op_fid3);
998 op_data->op_fid2 = op_data->op_fid3;
999 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1000 ll_finish_md_op_data(op_data);
1002 ll_update_times(request, dir);
1003 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1006 ptlrpc_req_finished(request);
1013 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1015 struct ptlrpc_request *request = NULL;
1016 struct md_op_data *op_data;
1020 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1021 namelen, name, dir->i_ino, dir->i_generation, dir);
1023 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1024 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1025 if (IS_ERR(op_data))
1026 RETURN(PTR_ERR(op_data));
1027 op_data->op_cli_flags |= CLI_RM_ENTRY;
1028 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1029 ll_finish_md_op_data(op_data);
1031 ll_update_times(request, dir);
1032 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1035 ptlrpc_req_finished(request);
1039 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1041 struct mdt_body *body;
1042 struct lov_mds_md *eadata;
1043 struct lov_stripe_md *lsm = NULL;
1044 struct obd_trans_info oti = { 0 };
1046 struct obd_capa *oc = NULL;
1050 /* req is swabbed so this is safe */
1051 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1052 if (!(body->valid & OBD_MD_FLEASIZE))
1055 if (body->eadatasize == 0) {
1056 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1057 GOTO(out, rc = -EPROTO);
1060 /* The MDS sent back the EA because we unlinked the last reference
1061 * to this file. Use this EA to unlink the objects on the OST.
1062 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1063 * check it is complete and sensible. */
1064 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1066 LASSERT(eadata != NULL);
1068 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1070 CERROR("obd_unpackmd: %d\n", rc);
1073 LASSERT(rc >= sizeof(*lsm));
1077 GOTO(out_free_memmd, rc = -ENOMEM);
1079 oa->o_oi = lsm->lsm_oi;
1080 oa->o_mode = body->mode & S_IFMT;
1081 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1083 if (body->valid & OBD_MD_FLCOOKIE) {
1084 oa->o_valid |= OBD_MD_FLCOOKIE;
1085 oti.oti_logcookies =
1086 req_capsule_server_sized_get(&request->rq_pill,
1088 sizeof(struct llog_cookie) *
1089 lsm->lsm_stripe_count);
1090 if (oti.oti_logcookies == NULL) {
1091 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1092 body->valid &= ~OBD_MD_FLCOOKIE;
1096 if (body->valid & OBD_MD_FLOSSCAPA) {
1097 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1099 GOTO(out_free_memmd, rc);
1102 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1103 ll_i2mdexp(dir), oc);
1106 CERROR("obd destroy objid "DOSTID" error %d\n",
1107 POSTID(&lsm->lsm_oi), rc);
1109 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1115 /* ll_unlink_generic() doesn't update the inode with the new link count.
1116 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1117 * is any lock existing. They will recycle dentries and inodes based upon locks
1119 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1120 struct dentry *dchild, struct qstr *name)
1122 struct ptlrpc_request *request = NULL;
1123 struct md_op_data *op_data;
1126 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1127 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1130 * XXX: unlink bind mountpoint maybe call to here,
1131 * just check it as vfs_unlink does.
1133 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1136 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1137 name->len, 0, LUSTRE_OPC_ANY, NULL);
1138 if (IS_ERR(op_data))
1139 RETURN(PTR_ERR(op_data));
1141 ll_get_child_fid(dir, name, &op_data->op_fid3);
1142 op_data->op_fid2 = op_data->op_fid3;
1143 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1144 ll_finish_md_op_data(op_data);
1148 ll_update_times(request, dir);
1149 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1151 rc = ll_objects_destroy(request, dir);
1153 ptlrpc_req_finished(request);
1157 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1158 struct dentry *src_dchild, struct qstr *src_name,
1159 struct inode *tgt, struct dentry *tgt_dparent,
1160 struct dentry *tgt_dchild, struct qstr *tgt_name)
1162 struct ptlrpc_request *request = NULL;
1163 struct ll_sb_info *sbi = ll_i2sbi(src);
1164 struct md_op_data *op_data;
1167 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1168 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1169 src->i_ino, src->i_generation, src, tgt_name->len,
1170 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1172 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1173 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1176 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1177 LUSTRE_OPC_ANY, NULL);
1178 if (IS_ERR(op_data))
1179 RETURN(PTR_ERR(op_data));
1181 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1182 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1183 err = md_rename(sbi->ll_md_exp, op_data,
1184 src_name->name, src_name->len,
1185 tgt_name->name, tgt_name->len, &request);
1186 ll_finish_md_op_data(op_data);
1188 ll_update_times(request, src);
1189 ll_update_times(request, tgt);
1190 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1191 err = ll_objects_destroy(request, src);
1194 ptlrpc_req_finished(request);
1199 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1202 return ll_mknod_generic(dir, &dchild->d_name, mode,
1203 old_encode_dev(rdev), dchild);
1206 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1208 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1211 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1213 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1216 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1218 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1221 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1222 const char *oldname)
1224 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1227 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1228 struct dentry *new_dentry)
1230 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1234 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1235 struct inode *new_dir, struct dentry *new_dentry)
1238 err = ll_rename_generic(old_dir, NULL,
1239 old_dentry, &old_dentry->d_name,
1240 new_dir, NULL, new_dentry,
1241 &new_dentry->d_name);
1243 d_move(old_dentry, new_dentry);
1248 struct inode_operations ll_dir_inode_operations = {
1250 .atomic_open = ll_atomic_open,
1251 .lookup = ll_lookup_nd,
1252 .create = ll_create_nd,
1253 /* We need all these non-raw things for NFSD, to not patch it. */
1254 .unlink = ll_unlink,
1257 .symlink = ll_symlink,
1259 .rename = ll_rename,
1260 .setattr = ll_setattr,
1261 .getattr = ll_getattr,
1262 .permission = ll_inode_permission,
1263 .setxattr = ll_setxattr,
1264 .getxattr = ll_getxattr,
1265 .listxattr = ll_listxattr,
1266 .removexattr = ll_removexattr,
1267 .get_acl = ll_get_acl,
1270 struct inode_operations ll_special_inode_operations = {
1271 .setattr = ll_setattr,
1272 .getattr = ll_getattr,
1273 .permission = ll_inode_permission,
1274 .setxattr = ll_setxattr,
1275 .getxattr = ll_getxattr,
1276 .listxattr = ll_listxattr,
1277 .removexattr = ll_removexattr,
1278 .get_acl = ll_get_acl,