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, 2013, 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 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: "DFID"(%p)\n",
170 PFID(&md->body->fid1), inode);
175 static void ll_invalidate_negative_children(struct inode *dir)
177 struct dentry *dentry, *tmp_subdir;
178 DECLARE_LL_D_HLIST_NODE_PTR(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_test_inode_by_fid(struct inode *inode, void *opaque)
200 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
203 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
204 void *data, int flag)
206 struct lustre_handle lockh;
211 case LDLM_CB_BLOCKING:
212 ldlm_lock2handle(lock, &lockh);
213 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
215 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
219 case LDLM_CB_CANCELING: {
220 struct inode *inode = ll_inode_from_resource_lock(lock);
221 __u64 bits = lock->l_policy_data.l_inodebits.bits;
223 /* Inode is set to lock->l_resource->lr_lvb_inode
224 * for mdc - bug 24555 */
225 LASSERT(lock->l_ast_data == NULL);
230 /* Invalidate all dentries associated with this inode */
231 LASSERT(ldlm_is_canceling(lock));
233 if (!fid_res_name_eq(ll_inode2fid(inode),
234 &lock->l_resource->lr_name)) {
235 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
236 PFID(ll_inode2fid(inode)), inode);
240 if (bits & MDS_INODELOCK_XATTR) {
241 ll_xattr_cache_destroy(inode);
242 bits &= ~MDS_INODELOCK_XATTR;
245 /* For OPEN locks we differentiate between lock modes
246 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
247 if (bits & MDS_INODELOCK_OPEN)
248 ll_have_md_lock(inode, &bits, lock->l_req_mode);
250 if (bits & MDS_INODELOCK_OPEN) {
253 switch (lock->l_req_mode) {
264 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
268 ll_md_real_close(inode, fmode);
270 bits &= ~MDS_INODELOCK_OPEN;
273 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
274 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
275 ll_have_md_lock(inode, &bits, LCK_MINMODE);
277 if (bits & MDS_INODELOCK_LAYOUT) {
278 struct cl_object_conf conf = {
279 .coc_opc = OBJECT_CONF_INVALIDATE,
283 rc = ll_layout_conf(inode, &conf);
285 CDEBUG(D_INODE, "cannot invalidate layout of "
287 PFID(ll_inode2fid(inode)), rc);
290 if (bits & MDS_INODELOCK_UPDATE) {
291 struct ll_inode_info *lli = ll_i2info(inode);
293 spin_lock(&lli->lli_lock);
294 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
295 spin_unlock(&lli->lli_lock);
298 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
299 struct ll_inode_info *lli = ll_i2info(inode);
301 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
302 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
303 lli, PFID(&lli->lli_pfid));
304 truncate_inode_pages(inode->i_mapping, 0);
306 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
307 struct inode *master_inode = NULL;
310 /* This is slave inode, since all of the child
311 * dentry is connected on the master inode, so
312 * we have to invalidate the negative children
314 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
315 PFID(ll_inode2fid(inode)),
316 PFID(&lli->lli_pfid));
318 hash = cl_fid_build_ino(&lli->lli_pfid,
319 ll_need_32bit_api(ll_i2sbi(inode)));
321 master_inode = ilookup5(inode->i_sb, hash,
322 ll_test_inode_by_fid,
323 (void *)&lli->lli_pfid);
324 if (master_inode != NULL &&
325 !IS_ERR(master_inode)) {
326 ll_invalidate_negative_children(
331 ll_invalidate_negative_children(inode);
335 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
336 inode->i_sb->s_root != NULL &&
337 inode != inode->i_sb->s_root->d_inode)
338 ll_invalidate_aliases(inode);
350 __u32 ll_i2suppgid(struct inode *i)
352 if (in_group_p(i->i_gid))
353 return (__u32)i->i_gid;
358 /* Pack the required supplementary groups into the supplied groups array.
359 * If we don't need to use the groups from the target inode(s) then we
360 * instead pack one or more groups from the user's supplementary group
361 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
362 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
369 LASSERT(suppgids != NULL);
371 suppgids[0] = ll_i2suppgid(i1);
374 suppgids[1] = ll_i2suppgid(i2);
379 for (i = 0; i < current_ngroups; i++) {
380 if (suppgids[0] == -1) {
381 if (current_groups[i] != suppgids[1])
382 suppgids[0] = current_groups[i];
385 if (suppgids[1] == -1) {
386 if (current_groups[i] != suppgids[0])
387 suppgids[1] = current_groups[i];
396 * try to reuse three types of dentry:
397 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
398 * by concurrent .revalidate).
399 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
400 * be cleared by others calling d_lustre_revalidate).
401 * 3. DISCONNECTED alias.
403 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
405 struct dentry *alias, *discon_alias, *invalid_alias;
406 DECLARE_LL_D_HLIST_NODE_PTR(p);
408 if (ll_d_hlist_empty(&inode->i_dentry))
411 discon_alias = invalid_alias = NULL;
413 ll_lock_dcache(inode);
414 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
415 LASSERT(alias != dentry);
417 spin_lock(&alias->d_lock);
418 if (alias->d_flags & DCACHE_DISCONNECTED)
419 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
420 discon_alias = alias;
421 else if (alias->d_parent == dentry->d_parent &&
422 alias->d_name.hash == dentry->d_name.hash &&
423 alias->d_name.len == dentry->d_name.len &&
424 memcmp(alias->d_name.name, dentry->d_name.name,
425 dentry->d_name.len) == 0)
426 invalid_alias = alias;
427 spin_unlock(&alias->d_lock);
432 alias = invalid_alias ?: discon_alias ?: NULL;
434 spin_lock(&alias->d_lock);
436 spin_unlock(&alias->d_lock);
438 ll_unlock_dcache(inode);
444 * Similar to d_splice_alias(), but lustre treats invalid alias
445 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
447 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
453 new = ll_find_alias(inode, de);
463 "Reuse dentry %p inode %p refc %d flags %#x\n",
464 new, new->d_inode, d_count(new), new->d_flags);
472 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
473 de, de->d_inode, d_count(de), de->d_flags);
477 int ll_lookup_it_finish(struct ptlrpc_request *request,
478 struct lookup_intent *it, void *data)
480 struct it_cb_data *icbd = data;
481 struct dentry **de = icbd->icbd_childp;
482 struct inode *parent = icbd->icbd_parent;
483 struct inode *inode = NULL;
488 /* NB 1 request reference will be taken away by ll_intent_lock()
490 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
491 it->d.lustre.it_disposition);
492 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
493 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
497 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
499 /* We used to query real size from OSTs here, but actually
500 this is not needed. For stat() calls size would be updated
501 from subsequent do_revalidate()->ll_inode_revalidate_it() in
503 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
504 Everybody else who needs correct file size would call
505 ll_glimpse_size or some equivalent themselves anyway.
506 Also see bug 7198. */
509 /* Only hash *de if it is unhashed (new dentry).
510 * Atoimc_open may passin hashed dentries for open.
512 if (d_unhashed(*de)) {
513 struct dentry *alias;
515 alias = ll_splice_alias(inode, *de);
517 RETURN(PTR_ERR(alias));
519 } else if (!it_disposition(it, DISP_LOOKUP_NEG) &&
520 !it_disposition(it, DISP_OPEN_CREATE)) {
521 /* With DISP_OPEN_CREATE dentry will
522 instantiated in ll_create_it. */
523 LASSERT((*de)->d_inode == NULL);
524 d_instantiate(*de, inode);
527 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
528 /* we have lookup look - unhide dentry */
529 if (bits & MDS_INODELOCK_LOOKUP)
530 d_lustre_revalidate(*de);
531 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
532 /* If file created on server, don't depend on parent UPDATE
533 * lock to unhide it. It is left hidden and next lookup can
534 * find it in ll_splice_alias.
536 /* Check that parent has UPDATE lock. */
537 struct lookup_intent parent_it = {
539 .d.lustre.it_lock_handle = 0 };
541 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
542 &ll_i2info(parent)->lli_fid, NULL)) {
543 d_lustre_revalidate(*de);
544 ll_intent_release(&parent_it);
551 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
552 struct lookup_intent *it, int lookup_flags)
554 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
555 struct dentry *save = dentry, *retval;
556 struct ptlrpc_request *req = NULL;
557 struct md_op_data *op_data;
558 struct it_cb_data icbd;
563 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
564 RETURN(ERR_PTR(-ENAMETOOLONG));
566 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
567 dentry->d_name.len, dentry->d_name.name,
568 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
570 if (d_mountpoint(dentry))
571 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
573 ll_frob_intent(&it, &lookup_it);
575 if (it->it_op == IT_GETATTR) {
576 rc = ll_statahead_enter(parent, &dentry, 0);
579 GOTO(out, retval = NULL);
580 GOTO(out, retval = dentry);
584 icbd.icbd_childp = &dentry;
585 icbd.icbd_parent = parent;
587 if (it->it_op & IT_CREAT)
588 opc = LUSTRE_OPC_CREATE;
590 opc = LUSTRE_OPC_ANY;
592 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
593 dentry->d_name.len, lookup_flags, opc,
596 RETURN((void *)op_data);
598 /* enforce umask if acl disabled or MDS doesn't support umask */
599 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
600 it->it_create_mode &= ~current_umask();
602 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
603 lookup_flags, &req, ll_md_blocking_ast, 0);
604 ll_finish_md_op_data(op_data);
606 GOTO(out, retval = ERR_PTR(rc));
608 rc = ll_lookup_it_finish(req, it, &icbd);
610 ll_intent_release(it);
611 GOTO(out, retval = ERR_PTR(rc));
614 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
615 !S_ISREG(dentry->d_inode->i_mode) &&
616 !S_ISDIR(dentry->d_inode->i_mode)) {
617 ll_release_openhandle(dentry, it);
619 ll_lookup_finish_locks(it, dentry);
622 GOTO(out, retval = NULL);
624 GOTO(out, retval = dentry);
627 ptlrpc_req_finished(req);
628 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
629 ll_statahead_mark(parent, dentry);
633 #ifdef HAVE_IOP_ATOMIC_OPEN
634 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
637 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
640 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
641 dentry->d_name.len, dentry->d_name.name,
642 PFID(ll_inode2fid(parent)), parent, flags);
644 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
645 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
648 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
652 de = ll_lookup_it(parent, dentry, itp, 0);
655 ll_intent_release(itp);
661 * For cached negative dentry and new dentry, handle lookup/create/open
664 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
665 struct file *file, unsigned open_flags,
666 umode_t mode, int *opened)
668 struct lookup_intent *it;
670 long long lookup_flags = LOOKUP_OPEN;
674 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
675 "open_flags %x, mode %x opened %d\n",
676 dentry->d_name.len, dentry->d_name.name,
677 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
679 OBD_ALLOC(it, sizeof(*it));
684 if (open_flags & O_CREAT) {
685 it->it_op |= IT_CREAT;
686 lookup_flags |= LOOKUP_CREATE;
688 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
689 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
691 /* Dentry added to dcache tree in ll_lookup_it */
692 de = ll_lookup_it(dir, dentry, it, lookup_flags);
699 if (it_disposition(it, DISP_OPEN_CREATE)) {
700 /* Dentry instantiated in ll_create_it. */
701 rc = ll_create_it(dir, dentry, mode, it);
703 /* We dget in ll_splice_alias. */
709 *opened |= FILE_CREATED;
711 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
713 if (S_ISFIFO(dentry->d_inode->i_mode)) {
714 /* We cannot call open here as it would
717 if (it_disposition(it, DISP_ENQ_OPEN_REF))
719 (struct ptlrpc_request *)
720 it->d.lustre.it_data);
721 rc = finish_no_open(file, de);
723 file->private_data = it;
724 rc = finish_open(file, dentry, NULL, opened);
725 /* We dget in ll_splice_alias. finish_open takes
726 * care of dget for fd open.
732 rc = finish_no_open(file, de);
737 ll_intent_release(it);
738 OBD_FREE(it, sizeof(*it));
743 #else /* !HAVE_IOP_ATOMIC_OPEN */
744 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
747 struct lookup_intent *it;
749 OBD_ALLOC(it, sizeof(*it));
751 return ERR_PTR(-ENOMEM);
753 if (lookup_flags & LOOKUP_OPEN) {
755 if (lookup_flags & LOOKUP_CREATE)
756 it->it_op |= IT_CREAT;
757 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
758 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
760 it->it_op = IT_GETATTR;
766 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
767 struct nameidata *nd)
772 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
773 struct lookup_intent *it;
775 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
776 it = ll_d2d(dentry)->lld_it;
777 ll_d2d(dentry)->lld_it = NULL;
779 if ((nd->flags & LOOKUP_CREATE) &&
780 !(nd->flags & LOOKUP_OPEN))
783 it = ll_convert_intent(&nd->intent.open, nd->flags);
785 RETURN((struct dentry *)it);
788 de = ll_lookup_it(parent, dentry, it, nd->flags);
791 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
792 if (dentry->d_inode &&
793 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
794 if (S_ISFIFO(dentry->d_inode->i_mode)) {
795 // We cannot call open here as it would
798 (struct ptlrpc_request *)
799 it->d.lustre.it_data);
803 nd->intent.open.file->private_data = it;
804 filp = lookup_instantiate_filp(nd,
810 de = (struct dentry *)filp;
813 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
814 // XXX This can only reliably work on assumption
815 // that there are NO hashed negative dentries.
816 ll_d2d(dentry)->lld_it = it;
817 it = NULL; /* Will be freed in ll_create_nd */
818 /* We absolutely depend on ll_create_nd to be
819 * called to not leak this intent and possible
820 * data attached to it */
825 ll_intent_release(it);
826 OBD_FREE(it, sizeof(*it));
829 de = ll_lookup_it(parent, dentry, NULL, 0);
834 #endif /* HAVE_IOP_ATOMIC_OPEN */
836 /* We depend on "mode" being set with the proper file type/umask by now */
837 static struct inode *ll_create_node(struct inode *dir, const char *name,
838 int namelen, const void *data, int datalen,
839 int mode, __u64 extra,
840 struct lookup_intent *it)
842 struct inode *inode = NULL;
843 struct ptlrpc_request *request = NULL;
844 struct ll_sb_info *sbi = ll_i2sbi(dir);
848 LASSERT(it && it->d.lustre.it_disposition);
850 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
851 request = it->d.lustre.it_data;
852 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
853 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
855 GOTO(out, inode = ERR_PTR(rc));
857 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
859 /* We asked for a lock on the directory, but were granted a
860 * lock on the inode. Since we finally have an inode pointer,
861 * stuff it in the lock. */
862 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
863 PFID(ll_inode2fid(inode)), inode);
864 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
867 ptlrpc_req_finished(request);
872 * By the time this is called, we already have created the directory cache
873 * entry for the new file, but it is so far negative - it has no inode.
875 * We defer creating the OBD object(s) until open, to keep the intent and
876 * non-intent code paths similar, and also because we do not have the MDS
877 * inode number before calling ll_create_node() (which is needed for LOV),
878 * so we would need to do yet another RPC to the MDS to store the LOV EA
879 * data on the MDS. If needed, we would pass the PACKED lmm as data and
880 * lmm_size in datalen (the MDS still has code which will handle that).
882 * If the create succeeds, we fill in the inode information
883 * with d_instantiate().
885 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
886 struct lookup_intent *it)
892 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
893 dentry->d_name.len, dentry->d_name.name,
894 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
896 rc = it_open_error(DISP_OPEN_CREATE, it);
900 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
901 NULL, 0, mode, 0, it);
903 RETURN(PTR_ERR(inode));
905 if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
906 ll_i2info(inode)->lli_volatile = true;
908 d_instantiate(dentry, inode);
912 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
914 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
918 if (body->valid & OBD_MD_FLMTIME &&
919 body->mtime > LTIME_S(inode->i_mtime)) {
920 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
921 "\n", PFID(ll_inode2fid(inode)),
922 LTIME_S(inode->i_mtime), body->mtime);
923 LTIME_S(inode->i_mtime) = body->mtime;
925 if (body->valid & OBD_MD_FLCTIME &&
926 body->ctime > LTIME_S(inode->i_ctime))
927 LTIME_S(inode->i_ctime) = body->ctime;
930 static int ll_new_node(struct inode *dir, struct qstr *name,
931 const char *tgt, int mode, int rdev,
932 struct dentry *dchild, __u32 opc)
934 struct ptlrpc_request *request = NULL;
935 struct md_op_data *op_data;
936 struct inode *inode = NULL;
937 struct ll_sb_info *sbi = ll_i2sbi(dir);
942 if (unlikely(tgt != NULL))
943 tgt_len = strlen(tgt) + 1;
945 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
946 name->len, 0, opc, NULL);
948 GOTO(err_exit, err = PTR_ERR(op_data));
950 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
951 current_fsuid(), current_fsgid(),
952 cfs_curproc_cap_pack(), rdev, &request);
953 ll_finish_md_op_data(op_data);
957 ll_update_times(request, dir);
960 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
964 d_instantiate(dchild, inode);
968 ptlrpc_req_finished(request);
973 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
974 unsigned rdev, struct dentry *dchild)
979 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
980 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
983 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
984 mode &= ~current_umask();
986 switch (mode & S_IFMT) {
988 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
994 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
1005 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1010 #ifdef HAVE_IOP_ATOMIC_OPEN
1012 * Plain create. Intent create is handled in atomic_open.
1014 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1015 umode_t mode, bool want_excl)
1019 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1020 "flags=%u, excl=%d\n", dentry->d_name.len,
1021 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1022 dir, mode, want_excl);
1024 rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1026 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1028 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1029 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1033 #else /* !HAVE_IOP_ATOMIC_OPEN */
1034 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1035 ll_umode_t mode, struct nameidata *nd)
1037 struct ll_dentry_data *lld = ll_d2d(dentry);
1038 struct lookup_intent *it = NULL;
1045 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1049 /* Was there an error? Propagate it! */
1050 if (it->d.lustre.it_status) {
1051 rc = it->d.lustre.it_status;
1055 rc = ll_create_it(dir, dentry, mode, it);
1056 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1059 nd->intent.open.file->private_data = it;
1060 filp = lookup_instantiate_filp(nd, dentry, NULL);
1066 ll_intent_release(it);
1067 OBD_FREE(it, sizeof(*it));
1070 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1074 #endif /* HAVE_IOP_ATOMIC_OPEN */
1076 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1077 const char *tgt, struct dentry *dchild)
1082 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1083 name->len, name->name, PFID(ll_inode2fid(dir)),
1086 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1087 0, dchild, LUSTRE_OPC_SYMLINK);
1090 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1095 static int ll_link_generic(struct inode *src, struct inode *dir,
1096 struct qstr *name, struct dentry *dchild)
1098 struct ll_sb_info *sbi = ll_i2sbi(dir);
1099 struct ptlrpc_request *request = NULL;
1100 struct md_op_data *op_data;
1104 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID
1105 "(%p), target=%.*s\n", PFID(ll_inode2fid(src)), src,
1106 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1108 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1109 0, LUSTRE_OPC_ANY, NULL);
1110 if (IS_ERR(op_data))
1111 RETURN(PTR_ERR(op_data));
1113 err = md_link(sbi->ll_md_exp, op_data, &request);
1114 ll_finish_md_op_data(op_data);
1118 ll_update_times(request, dir);
1119 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1122 ptlrpc_req_finished(request);
1126 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1127 int mode, struct dentry *dchild)
1133 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1134 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1136 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1137 mode &= ~current_umask();
1138 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1139 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1142 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1147 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1148 struct dentry *dchild, struct qstr *name)
1150 struct ptlrpc_request *request = NULL;
1151 struct md_op_data *op_data;
1155 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1156 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1158 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1161 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1162 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1163 if (IS_ERR(op_data))
1164 RETURN(PTR_ERR(op_data));
1166 if (dchild != NULL && dchild->d_inode != NULL)
1167 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1168 op_data->op_fid2 = op_data->op_fid3;
1169 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1170 ll_finish_md_op_data(op_data);
1172 ll_update_times(request, dir);
1173 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1176 ptlrpc_req_finished(request);
1183 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1185 struct ptlrpc_request *request = NULL;
1186 struct md_op_data *op_data;
1190 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1191 namelen, name, PFID(ll_inode2fid(dir)), dir);
1193 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1194 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1195 if (IS_ERR(op_data))
1196 RETURN(PTR_ERR(op_data));
1197 op_data->op_cli_flags |= CLI_RM_ENTRY;
1198 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1199 ll_finish_md_op_data(op_data);
1201 ll_update_times(request, dir);
1202 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1205 ptlrpc_req_finished(request);
1209 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1211 struct mdt_body *body;
1212 struct lov_mds_md *eadata;
1213 struct lov_stripe_md *lsm = NULL;
1214 struct obd_trans_info oti = { 0 };
1216 struct obd_capa *oc = NULL;
1220 /* req is swabbed so this is safe */
1221 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1222 if (!(body->valid & OBD_MD_FLEASIZE))
1225 if (body->eadatasize == 0) {
1226 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1227 GOTO(out, rc = -EPROTO);
1230 /* The MDS sent back the EA because we unlinked the last reference
1231 * to this file. Use this EA to unlink the objects on the OST.
1232 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1233 * check it is complete and sensible. */
1234 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1236 LASSERT(eadata != NULL);
1238 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1240 CERROR("obd_unpackmd: %d\n", rc);
1243 LASSERT(rc >= sizeof(*lsm));
1247 GOTO(out_free_memmd, rc = -ENOMEM);
1249 oa->o_oi = lsm->lsm_oi;
1250 oa->o_mode = body->mode & S_IFMT;
1251 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1253 if (body->valid & OBD_MD_FLCOOKIE) {
1254 oa->o_valid |= OBD_MD_FLCOOKIE;
1255 oti.oti_logcookies =
1256 req_capsule_server_sized_get(&request->rq_pill,
1258 sizeof(struct llog_cookie) *
1259 lsm->lsm_stripe_count);
1260 if (oti.oti_logcookies == NULL) {
1261 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1262 body->valid &= ~OBD_MD_FLCOOKIE;
1266 if (body->valid & OBD_MD_FLOSSCAPA) {
1267 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1269 GOTO(out_free_memmd, rc);
1272 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1273 ll_i2mdexp(dir), oc);
1276 CERROR("obd destroy objid "DOSTID" error %d\n",
1277 POSTID(&lsm->lsm_oi), rc);
1279 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1285 /* ll_unlink_generic() doesn't update the inode with the new link count.
1286 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1287 * is any lock existing. They will recycle dentries and inodes based upon locks
1289 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1290 struct dentry *dchild, struct qstr *name)
1292 struct ptlrpc_request *request = NULL;
1293 struct md_op_data *op_data;
1296 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1297 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1300 * XXX: unlink bind mountpoint maybe call to here,
1301 * just check it as vfs_unlink does.
1303 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1306 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1307 name->len, 0, LUSTRE_OPC_ANY, NULL);
1308 if (IS_ERR(op_data))
1309 RETURN(PTR_ERR(op_data));
1311 if (dchild != NULL && dchild->d_inode != NULL)
1312 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1314 op_data->op_fid2 = op_data->op_fid3;
1315 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1316 ll_finish_md_op_data(op_data);
1320 ll_update_times(request, dir);
1321 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1323 rc = ll_objects_destroy(request, dir);
1325 ptlrpc_req_finished(request);
1329 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1330 struct dentry *src_dchild, struct qstr *src_name,
1331 struct inode *tgt, struct dentry *tgt_dparent,
1332 struct dentry *tgt_dchild, struct qstr *tgt_name)
1334 struct ptlrpc_request *request = NULL;
1335 struct ll_sb_info *sbi = ll_i2sbi(src);
1336 struct md_op_data *op_data;
1339 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1340 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1341 src_name->len, src_name->name,
1342 PFID(ll_inode2fid(src)), src, tgt_name->len,
1343 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1345 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1346 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1349 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1350 LUSTRE_OPC_ANY, NULL);
1351 if (IS_ERR(op_data))
1352 RETURN(PTR_ERR(op_data));
1354 if (src_dchild != NULL && src_dchild->d_inode != NULL)
1355 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1356 if (tgt_dchild != NULL && tgt_dchild->d_inode != NULL)
1357 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1359 err = md_rename(sbi->ll_md_exp, op_data,
1360 src_name->name, src_name->len,
1361 tgt_name->name, tgt_name->len, &request);
1362 ll_finish_md_op_data(op_data);
1364 ll_update_times(request, src);
1365 ll_update_times(request, tgt);
1366 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1367 err = ll_objects_destroy(request, src);
1370 ptlrpc_req_finished(request);
1375 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1378 return ll_mknod_generic(dir, &dchild->d_name, mode,
1379 old_encode_dev(rdev), dchild);
1382 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1384 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1387 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1389 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1392 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1394 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1397 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1398 const char *oldname)
1400 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1403 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1404 struct dentry *new_dentry)
1406 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1410 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1411 struct inode *new_dir, struct dentry *new_dentry)
1414 err = ll_rename_generic(old_dir, NULL,
1415 old_dentry, &old_dentry->d_name,
1416 new_dir, NULL, new_dentry,
1417 &new_dentry->d_name);
1419 d_move(old_dentry, new_dentry);
1423 struct inode_operations ll_dir_inode_operations = {
1425 #ifdef HAVE_IOP_ATOMIC_OPEN
1426 .atomic_open = ll_atomic_open,
1428 .lookup = ll_lookup_nd,
1429 .create = ll_create_nd,
1430 /* We need all these non-raw things for NFSD, to not patch it. */
1431 .unlink = ll_unlink,
1434 .symlink = ll_symlink,
1436 .rename = ll_rename,
1437 .setattr = ll_setattr,
1438 .getattr = ll_getattr,
1439 .permission = ll_inode_permission,
1440 .setxattr = ll_setxattr,
1441 .getxattr = ll_getxattr,
1442 .listxattr = ll_listxattr,
1443 .removexattr = ll_removexattr,
1444 #ifdef HAVE_IOP_GET_ACL
1445 .get_acl = ll_get_acl,
1449 struct inode_operations ll_special_inode_operations = {
1450 .setattr = ll_setattr,
1451 .getattr = ll_getattr,
1452 .permission = ll_inode_permission,
1453 .setxattr = ll_setxattr,
1454 .getxattr = ll_getxattr,
1455 .listxattr = ll_listxattr,
1456 .removexattr = ll_removexattr,
1457 #ifdef HAVE_IOP_GET_ACL
1458 .get_acl = ll_get_acl,