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 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 /* called from iget5_locked->find_inode() under inode_lock spinlock */
79 static int ll_test_inode(struct inode *inode, void *opaque)
81 struct ll_inode_info *lli = ll_i2info(inode);
82 struct lustre_md *md = opaque;
84 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
85 CERROR("MDS body missing FID\n");
89 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
95 static int ll_set_inode(struct inode *inode, void *opaque)
97 struct ll_inode_info *lli = ll_i2info(inode);
98 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
100 if (unlikely(!(body->valid & OBD_MD_FLID))) {
101 CERROR("MDS body missing FID\n");
105 lli->lli_fid = body->fid1;
106 if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
107 CERROR("Can not initialize inode "DFID" without object type: "
108 "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
112 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
113 if (unlikely(inode->i_mode == 0)) {
114 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
125 * Get an inode by inode number(@hash), which is already instantiated by
126 * the intent lookup).
128 struct inode *ll_iget(struct super_block *sb, ino_t hash,
129 struct lustre_md *md)
137 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
139 RETURN(ERR_PTR(-ENOMEM));
141 if (inode->i_state & I_NEW) {
142 rc = ll_read_inode2(inode, md);
143 if (rc == 0 && S_ISREG(inode->i_mode) &&
144 ll_i2info(inode)->lli_clob == NULL) {
145 CDEBUG(D_INODE, "%s: apply lsm %p to inode "DFID"\n",
146 ll_get_fsname(sb, NULL, 0), md->lsm,
147 PFID(ll_inode2fid(inode)));
148 rc = cl_file_inode_init(inode, md);
151 make_bad_inode(inode);
152 unlock_new_inode(inode);
156 unlock_new_inode(inode);
158 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
159 rc = ll_update_inode(inode, md);
160 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
161 PFID(&md->body->fid1), inode, rc);
163 make_bad_inode(inode);
172 static void ll_invalidate_negative_children(struct inode *dir)
174 struct dentry *dentry, *tmp_subdir;
175 DECLARE_LL_D_HLIST_NODE_PTR(p);
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;
183 list_for_each_entry_safe(child, tmp_subdir,
186 if (child->d_inode == NULL)
187 d_lustre_invalidate(child, 1);
190 spin_unlock(&dentry->d_lock);
192 ll_unlock_dcache(dir);
195 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
197 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
200 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
201 void *data, int flag)
203 struct lustre_handle lockh;
208 case LDLM_CB_BLOCKING:
209 ldlm_lock2handle(lock, &lockh);
210 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
212 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
216 case LDLM_CB_CANCELING: {
217 struct inode *inode = ll_inode_from_resource_lock(lock);
218 __u64 bits = lock->l_policy_data.l_inodebits.bits;
220 /* Inode is set to lock->l_resource->lr_lvb_inode
221 * for mdc - bug 24555 */
222 LASSERT(lock->l_ast_data == NULL);
227 /* Invalidate all dentries associated with this inode */
228 LASSERT(ldlm_is_canceling(lock));
230 if (!fid_res_name_eq(ll_inode2fid(inode),
231 &lock->l_resource->lr_name)) {
232 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
233 PFID(ll_inode2fid(inode)), inode);
237 if (bits & MDS_INODELOCK_XATTR) {
238 ll_xattr_cache_destroy(inode);
239 bits &= ~MDS_INODELOCK_XATTR;
242 /* For OPEN locks we differentiate between lock modes
243 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
244 if (bits & MDS_INODELOCK_OPEN)
245 ll_have_md_lock(inode, &bits, lock->l_req_mode);
247 if (bits & MDS_INODELOCK_OPEN) {
250 switch (lock->l_req_mode) {
261 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
265 ll_md_real_close(inode, fmode);
267 bits &= ~MDS_INODELOCK_OPEN;
270 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
271 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
272 ll_have_md_lock(inode, &bits, LCK_MINMODE);
274 if (bits & MDS_INODELOCK_LAYOUT) {
275 struct cl_object_conf conf = {
276 .coc_opc = OBJECT_CONF_INVALIDATE,
280 rc = ll_layout_conf(inode, &conf);
282 CDEBUG(D_INODE, "cannot invalidate layout of "
284 PFID(ll_inode2fid(inode)), rc);
287 if (bits & MDS_INODELOCK_UPDATE) {
288 struct ll_inode_info *lli = ll_i2info(inode);
290 spin_lock(&lli->lli_lock);
291 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
292 spin_unlock(&lli->lli_lock);
295 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
296 struct ll_inode_info *lli = ll_i2info(inode);
298 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
299 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
300 lli, PFID(&lli->lli_pfid));
301 truncate_inode_pages(inode->i_mapping, 0);
303 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
304 struct inode *master_inode = NULL;
307 /* This is slave inode, since all of the child
308 * dentry is connected on the master inode, so
309 * we have to invalidate the negative children
311 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
312 PFID(ll_inode2fid(inode)),
313 PFID(&lli->lli_pfid));
315 hash = cl_fid_build_ino(&lli->lli_pfid,
316 ll_need_32bit_api(ll_i2sbi(inode)));
318 master_inode = ilookup5(inode->i_sb, hash,
319 ll_test_inode_by_fid,
320 (void *)&lli->lli_pfid);
321 if (master_inode != NULL &&
322 !IS_ERR(master_inode)) {
323 ll_invalidate_negative_children(
328 ll_invalidate_negative_children(inode);
332 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
333 inode->i_sb->s_root != NULL &&
334 inode != inode->i_sb->s_root->d_inode)
335 ll_invalidate_aliases(inode);
347 __u32 ll_i2suppgid(struct inode *i)
349 if (in_group_p(i->i_gid))
350 return (__u32)from_kgid(&init_user_ns, i->i_gid);
352 return (__u32) __kgid_val(INVALID_GID);
355 /* Pack the required supplementary groups into the supplied groups array.
356 * If we don't need to use the groups from the target inode(s) then we
357 * instead pack one or more groups from the user's supplementary group
358 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
359 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
362 LASSERT(suppgids != NULL);
364 suppgids[0] = ll_i2suppgid(i1);
367 suppgids[1] = ll_i2suppgid(i2);
373 * try to reuse three types of dentry:
374 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
375 * by concurrent .revalidate).
376 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
377 * be cleared by others calling d_lustre_revalidate).
378 * 3. DISCONNECTED alias.
380 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
382 struct dentry *alias, *discon_alias, *invalid_alias;
383 DECLARE_LL_D_HLIST_NODE_PTR(p);
385 if (ll_d_hlist_empty(&inode->i_dentry))
388 discon_alias = invalid_alias = NULL;
390 ll_lock_dcache(inode);
391 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
392 LASSERT(alias != dentry);
394 spin_lock(&alias->d_lock);
395 if (alias->d_flags & DCACHE_DISCONNECTED)
396 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
397 discon_alias = alias;
398 else if (alias->d_parent == dentry->d_parent &&
399 alias->d_name.hash == dentry->d_name.hash &&
400 alias->d_name.len == dentry->d_name.len &&
401 memcmp(alias->d_name.name, dentry->d_name.name,
402 dentry->d_name.len) == 0)
403 invalid_alias = alias;
404 spin_unlock(&alias->d_lock);
409 alias = invalid_alias ?: discon_alias ?: NULL;
411 spin_lock(&alias->d_lock);
413 spin_unlock(&alias->d_lock);
415 ll_unlock_dcache(inode);
421 * Similar to d_splice_alias(), but lustre treats invalid alias
422 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
424 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
430 new = ll_find_alias(inode, de);
440 "Reuse dentry %p inode %p refc %d flags %#x\n",
441 new, new->d_inode, d_count(new), new->d_flags);
449 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
450 de, de->d_inode, d_count(de), de->d_flags);
454 static int ll_lookup_it_finish(struct ptlrpc_request *request,
455 struct lookup_intent *it,
456 struct inode *parent, struct dentry **de)
458 struct inode *inode = NULL;
463 /* NB 1 request reference will be taken away by ll_intent_lock()
465 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
466 it->d.lustre.it_disposition);
467 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
468 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
472 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
474 /* We used to query real size from OSTs here, but actually
475 this is not needed. For stat() calls size would be updated
476 from subsequent do_revalidate()->ll_inode_revalidate_it() in
478 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
479 Everybody else who needs correct file size would call
480 ll_glimpse_size or some equivalent themselves anyway.
481 Also see bug 7198. */
484 /* Only hash *de if it is unhashed (new dentry).
485 * Atoimc_open may passin hashed dentries for open.
487 if (d_unhashed(*de)) {
488 struct dentry *alias;
490 alias = ll_splice_alias(inode, *de);
492 RETURN(PTR_ERR(alias));
494 } else if (!it_disposition(it, DISP_LOOKUP_NEG) &&
495 !it_disposition(it, DISP_OPEN_CREATE)) {
496 /* With DISP_OPEN_CREATE dentry will
497 instantiated in ll_create_it. */
498 LASSERT((*de)->d_inode == NULL);
499 d_instantiate(*de, inode);
502 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
503 /* we have lookup look - unhide dentry */
504 if (bits & MDS_INODELOCK_LOOKUP)
505 d_lustre_revalidate(*de);
506 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
507 /* If file created on server, don't depend on parent UPDATE
508 * lock to unhide it. It is left hidden and next lookup can
509 * find it in ll_splice_alias.
511 /* Check that parent has UPDATE lock. */
512 struct lookup_intent parent_it = {
514 .d.lustre.it_lock_handle = 0 };
516 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
517 &ll_i2info(parent)->lli_fid, NULL)) {
518 d_lustre_revalidate(*de);
519 ll_intent_release(&parent_it);
526 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
527 struct lookup_intent *it, int lookup_flags)
529 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
530 struct dentry *save = dentry, *retval;
531 struct ptlrpc_request *req = NULL;
532 struct md_op_data *op_data;
537 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
538 RETURN(ERR_PTR(-ENAMETOOLONG));
540 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
541 dentry->d_name.len, dentry->d_name.name,
542 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
544 if (d_mountpoint(dentry))
545 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
547 if (it == NULL || it->it_op == IT_GETXATTR)
550 if (it->it_op == IT_GETATTR) {
551 rc = ll_statahead_enter(parent, &dentry, 0);
554 GOTO(out, retval = NULL);
555 GOTO(out, retval = dentry);
559 if (it->it_op & IT_CREAT)
560 opc = LUSTRE_OPC_CREATE;
562 opc = LUSTRE_OPC_ANY;
564 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
565 dentry->d_name.len, lookup_flags, opc,
568 RETURN((void *)op_data);
570 /* enforce umask if acl disabled or MDS doesn't support umask */
571 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
572 it->it_create_mode &= ~current_umask();
574 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
575 &ll_md_blocking_ast, 0);
576 ll_finish_md_op_data(op_data);
578 GOTO(out, retval = ERR_PTR(rc));
580 rc = ll_lookup_it_finish(req, it, parent, &dentry);
582 ll_intent_release(it);
583 GOTO(out, retval = ERR_PTR(rc));
586 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
587 !S_ISREG(dentry->d_inode->i_mode) &&
588 !S_ISDIR(dentry->d_inode->i_mode)) {
589 ll_release_openhandle(dentry, it);
591 ll_lookup_finish_locks(it, dentry);
594 GOTO(out, retval = NULL);
596 GOTO(out, retval = dentry);
599 ptlrpc_req_finished(req);
600 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
601 ll_statahead_mark(parent, dentry);
605 #ifdef HAVE_IOP_ATOMIC_OPEN
606 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
609 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
612 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
613 dentry->d_name.len, dentry->d_name.name,
614 PFID(ll_inode2fid(parent)), parent, flags);
616 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
617 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
620 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
624 de = ll_lookup_it(parent, dentry, itp, 0);
627 ll_intent_release(itp);
633 * For cached negative dentry and new dentry, handle lookup/create/open
636 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
637 struct file *file, unsigned open_flags,
638 umode_t mode, int *opened)
640 struct lookup_intent *it;
642 long long lookup_flags = LOOKUP_OPEN;
646 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
647 "open_flags %x, mode %x opened %d\n",
648 dentry->d_name.len, dentry->d_name.name,
649 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
651 OBD_ALLOC(it, sizeof(*it));
656 if (open_flags & O_CREAT) {
657 it->it_op |= IT_CREAT;
658 lookup_flags |= LOOKUP_CREATE;
660 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
661 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
663 /* Dentry added to dcache tree in ll_lookup_it */
664 de = ll_lookup_it(dir, dentry, it, lookup_flags);
671 if (it_disposition(it, DISP_OPEN_CREATE)) {
672 /* Dentry instantiated in ll_create_it. */
673 rc = ll_create_it(dir, dentry, mode, it);
675 /* We dget in ll_splice_alias. */
681 *opened |= FILE_CREATED;
683 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
685 if (S_ISFIFO(dentry->d_inode->i_mode)) {
686 /* We cannot call open here as it would
689 if (it_disposition(it, DISP_ENQ_OPEN_REF))
691 (struct ptlrpc_request *)
692 it->d.lustre.it_data);
693 rc = finish_no_open(file, de);
695 file->private_data = it;
696 rc = finish_open(file, dentry, NULL, opened);
697 /* We dget in ll_splice_alias. finish_open takes
698 * care of dget for fd open.
704 rc = finish_no_open(file, de);
709 ll_intent_release(it);
710 OBD_FREE(it, sizeof(*it));
715 #else /* !HAVE_IOP_ATOMIC_OPEN */
716 static struct lookup_intent *
717 ll_convert_intent(struct open_intent *oit, int lookup_flags)
719 struct lookup_intent *it;
721 OBD_ALLOC(it, sizeof(*it));
723 return ERR_PTR(-ENOMEM);
725 if (lookup_flags & LOOKUP_OPEN) {
727 if (lookup_flags & LOOKUP_CREATE)
728 it->it_op |= IT_CREAT;
729 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
730 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
732 it->it_op = IT_GETATTR;
738 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
739 struct nameidata *nd)
744 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
745 struct lookup_intent *it;
747 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
748 it = ll_d2d(dentry)->lld_it;
749 ll_d2d(dentry)->lld_it = NULL;
751 if ((nd->flags & LOOKUP_CREATE) &&
752 !(nd->flags & LOOKUP_OPEN))
755 it = ll_convert_intent(&nd->intent.open, nd->flags);
757 RETURN((struct dentry *)it);
760 de = ll_lookup_it(parent, dentry, it, nd->flags);
763 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
764 if (dentry->d_inode &&
765 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
766 if (S_ISFIFO(dentry->d_inode->i_mode)) {
767 // We cannot call open here as it would
770 (struct ptlrpc_request *)
771 it->d.lustre.it_data);
775 nd->intent.open.file->private_data = it;
776 filp = lookup_instantiate_filp(nd,
782 de = (struct dentry *)filp;
785 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
786 // XXX This can only reliably work on assumption
787 // that there are NO hashed negative dentries.
788 ll_d2d(dentry)->lld_it = it;
789 it = NULL; /* Will be freed in ll_create_nd */
790 /* We absolutely depend on ll_create_nd to be
791 * called to not leak this intent and possible
792 * data attached to it */
797 ll_intent_release(it);
798 OBD_FREE(it, sizeof(*it));
801 de = ll_lookup_it(parent, dentry, NULL, 0);
806 #endif /* HAVE_IOP_ATOMIC_OPEN */
808 /* We depend on "mode" being set with the proper file type/umask by now */
809 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
811 struct inode *inode = NULL;
812 struct ptlrpc_request *request = NULL;
813 struct ll_sb_info *sbi = ll_i2sbi(dir);
817 LASSERT(it && it->d.lustre.it_disposition);
819 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
820 request = it->d.lustre.it_data;
821 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
822 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
824 GOTO(out, inode = ERR_PTR(rc));
826 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
828 /* We asked for a lock on the directory, but were granted a
829 * lock on the inode. Since we finally have an inode pointer,
830 * stuff it in the lock. */
831 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
832 PFID(ll_inode2fid(inode)), inode);
833 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
836 ptlrpc_req_finished(request);
841 * By the time this is called, we already have created the directory cache
842 * entry for the new file, but it is so far negative - it has no inode.
844 * We defer creating the OBD object(s) until open, to keep the intent and
845 * non-intent code paths similar, and also because we do not have the MDS
846 * inode number before calling ll_create_node() (which is needed for LOV),
847 * so we would need to do yet another RPC to the MDS to store the LOV EA
848 * data on the MDS. If needed, we would pass the PACKED lmm as data and
849 * lmm_size in datalen (the MDS still has code which will handle that).
851 * If the create succeeds, we fill in the inode information
852 * with d_instantiate().
854 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
855 struct lookup_intent *it)
861 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
862 dentry->d_name.len, dentry->d_name.name,
863 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
865 rc = it_open_error(DISP_OPEN_CREATE, it);
869 inode = ll_create_node(dir, it);
871 RETURN(PTR_ERR(inode));
873 d_instantiate(dentry, inode);
877 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
879 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
883 if (body->valid & OBD_MD_FLMTIME &&
884 body->mtime > LTIME_S(inode->i_mtime)) {
885 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
886 "\n", PFID(ll_inode2fid(inode)),
887 LTIME_S(inode->i_mtime), body->mtime);
888 LTIME_S(inode->i_mtime) = body->mtime;
890 if (body->valid & OBD_MD_FLCTIME &&
891 body->ctime > LTIME_S(inode->i_ctime))
892 LTIME_S(inode->i_ctime) = body->ctime;
895 static int ll_new_node(struct inode *dir, struct qstr *name,
896 const char *tgt, int mode, int rdev,
897 struct dentry *dchild, __u32 opc)
899 struct ptlrpc_request *request = NULL;
900 struct md_op_data *op_data;
901 struct inode *inode = NULL;
902 struct ll_sb_info *sbi = ll_i2sbi(dir);
907 if (unlikely(tgt != NULL))
908 tgt_len = strlen(tgt) + 1;
910 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
911 name->len, 0, opc, NULL);
913 GOTO(err_exit, err = PTR_ERR(op_data));
915 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
916 from_kuid(&init_user_ns, current_fsuid()),
917 from_kgid(&init_user_ns, current_fsgid()),
918 cfs_curproc_cap_pack(), rdev, &request);
919 ll_finish_md_op_data(op_data);
923 ll_update_times(request, dir);
926 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
930 d_instantiate(dchild, inode);
934 ptlrpc_req_finished(request);
939 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
940 unsigned rdev, struct dentry *dchild)
945 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
946 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
949 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
950 mode &= ~current_umask();
952 switch (mode & S_IFMT) {
954 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
960 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
971 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
976 #ifdef HAVE_IOP_ATOMIC_OPEN
978 * Plain create. Intent create is handled in atomic_open.
980 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
981 umode_t mode, bool want_excl)
985 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
986 "flags=%u, excl=%d\n", dentry->d_name.len,
987 dentry->d_name.name, PFID(ll_inode2fid(dir)),
988 dir, mode, want_excl);
990 rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
992 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
994 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
995 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
999 #else /* !HAVE_IOP_ATOMIC_OPEN */
1000 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1001 ll_umode_t mode, struct nameidata *nd)
1003 struct ll_dentry_data *lld = ll_d2d(dentry);
1004 struct lookup_intent *it = NULL;
1011 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1015 /* Was there an error? Propagate it! */
1016 if (it->d.lustre.it_status) {
1017 rc = it->d.lustre.it_status;
1021 rc = ll_create_it(dir, dentry, mode, it);
1022 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1025 nd->intent.open.file->private_data = it;
1026 filp = lookup_instantiate_filp(nd, dentry, NULL);
1032 ll_intent_release(it);
1033 OBD_FREE(it, sizeof(*it));
1036 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1040 #endif /* HAVE_IOP_ATOMIC_OPEN */
1042 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1043 const char *tgt, struct dentry *dchild)
1048 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1049 name->len, name->name, PFID(ll_inode2fid(dir)),
1052 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1053 0, dchild, LUSTRE_OPC_SYMLINK);
1056 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1061 static int ll_link_generic(struct inode *src, struct inode *dir,
1062 struct qstr *name, struct dentry *dchild)
1064 struct ll_sb_info *sbi = ll_i2sbi(dir);
1065 struct ptlrpc_request *request = NULL;
1066 struct md_op_data *op_data;
1070 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID
1071 "(%p), target=%.*s\n", PFID(ll_inode2fid(src)), src,
1072 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1074 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1075 0, LUSTRE_OPC_ANY, NULL);
1076 if (IS_ERR(op_data))
1077 RETURN(PTR_ERR(op_data));
1079 err = md_link(sbi->ll_md_exp, op_data, &request);
1080 ll_finish_md_op_data(op_data);
1084 ll_update_times(request, dir);
1085 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1088 ptlrpc_req_finished(request);
1092 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1093 int mode, struct dentry *dchild)
1099 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1100 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1102 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1103 mode &= ~current_umask();
1104 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1105 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1108 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1113 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1114 struct dentry *dchild, struct qstr *name)
1116 struct ptlrpc_request *request = NULL;
1117 struct md_op_data *op_data;
1121 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1122 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1124 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1127 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1128 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1129 if (IS_ERR(op_data))
1130 RETURN(PTR_ERR(op_data));
1132 if (dchild != NULL && dchild->d_inode != NULL)
1133 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1134 op_data->op_fid2 = op_data->op_fid3;
1135 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1136 ll_finish_md_op_data(op_data);
1138 ll_update_times(request, dir);
1139 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1142 ptlrpc_req_finished(request);
1149 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1151 struct ptlrpc_request *request = NULL;
1152 struct md_op_data *op_data;
1156 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1157 namelen, name, PFID(ll_inode2fid(dir)), dir);
1159 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1160 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1161 if (IS_ERR(op_data))
1162 RETURN(PTR_ERR(op_data));
1163 op_data->op_cli_flags |= CLI_RM_ENTRY;
1164 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1165 ll_finish_md_op_data(op_data);
1167 ll_update_times(request, dir);
1168 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1171 ptlrpc_req_finished(request);
1175 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1177 struct mdt_body *body;
1178 struct lov_mds_md *eadata;
1179 struct lov_stripe_md *lsm = NULL;
1180 struct obd_trans_info oti = { 0 };
1182 struct obd_capa *oc = NULL;
1186 /* req is swabbed so this is safe */
1187 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1188 if (!(body->valid & OBD_MD_FLEASIZE))
1191 if (body->eadatasize == 0) {
1192 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1193 GOTO(out, rc = -EPROTO);
1196 /* The MDS sent back the EA because we unlinked the last reference
1197 * to this file. Use this EA to unlink the objects on the OST.
1198 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1199 * check it is complete and sensible. */
1200 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1202 LASSERT(eadata != NULL);
1204 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1206 CERROR("obd_unpackmd: %d\n", rc);
1209 LASSERT(rc >= sizeof(*lsm));
1213 GOTO(out_free_memmd, rc = -ENOMEM);
1215 oa->o_oi = lsm->lsm_oi;
1216 oa->o_mode = body->mode & S_IFMT;
1217 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1219 if (body->valid & OBD_MD_FLCOOKIE) {
1220 oa->o_valid |= OBD_MD_FLCOOKIE;
1221 oti.oti_logcookies =
1222 req_capsule_server_sized_get(&request->rq_pill,
1224 sizeof(struct llog_cookie) *
1225 lsm->lsm_stripe_count);
1226 if (oti.oti_logcookies == NULL) {
1227 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1228 body->valid &= ~OBD_MD_FLCOOKIE;
1232 if (body->valid & OBD_MD_FLOSSCAPA) {
1233 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1235 GOTO(out_free_memmd, rc);
1238 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1239 ll_i2mdexp(dir), oc);
1242 CERROR("obd destroy objid "DOSTID" error %d\n",
1243 POSTID(&lsm->lsm_oi), rc);
1245 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1251 /* ll_unlink_generic() doesn't update the inode with the new link count.
1252 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1253 * is any lock existing. They will recycle dentries and inodes based upon locks
1255 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1256 struct dentry *dchild, struct qstr *name)
1258 struct ptlrpc_request *request = NULL;
1259 struct md_op_data *op_data;
1262 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1263 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1266 * XXX: unlink bind mountpoint maybe call to here,
1267 * just check it as vfs_unlink does.
1269 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1272 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1273 name->len, 0, LUSTRE_OPC_ANY, NULL);
1274 if (IS_ERR(op_data))
1275 RETURN(PTR_ERR(op_data));
1277 if (dchild != NULL && dchild->d_inode != NULL)
1278 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1280 op_data->op_fid2 = op_data->op_fid3;
1281 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1282 ll_finish_md_op_data(op_data);
1286 ll_update_times(request, dir);
1287 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1289 rc = ll_objects_destroy(request, dir);
1291 ptlrpc_req_finished(request);
1295 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1296 struct dentry *src_dchild, struct qstr *src_name,
1297 struct inode *tgt, struct dentry *tgt_dparent,
1298 struct dentry *tgt_dchild, struct qstr *tgt_name)
1300 struct ptlrpc_request *request = NULL;
1301 struct ll_sb_info *sbi = ll_i2sbi(src);
1302 struct md_op_data *op_data;
1305 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1306 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1307 src_name->len, src_name->name,
1308 PFID(ll_inode2fid(src)), src, tgt_name->len,
1309 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1311 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1312 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1315 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1316 LUSTRE_OPC_ANY, NULL);
1317 if (IS_ERR(op_data))
1318 RETURN(PTR_ERR(op_data));
1320 if (src_dchild != NULL && src_dchild->d_inode != NULL)
1321 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1322 if (tgt_dchild != NULL && tgt_dchild->d_inode != NULL)
1323 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1325 err = md_rename(sbi->ll_md_exp, op_data,
1326 src_name->name, src_name->len,
1327 tgt_name->name, tgt_name->len, &request);
1328 ll_finish_md_op_data(op_data);
1330 ll_update_times(request, src);
1331 ll_update_times(request, tgt);
1332 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1333 err = ll_objects_destroy(request, src);
1336 ptlrpc_req_finished(request);
1341 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1344 return ll_mknod_generic(dir, &dchild->d_name, mode,
1345 old_encode_dev(rdev), dchild);
1348 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1350 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1353 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1355 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1358 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1360 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1363 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1364 const char *oldname)
1366 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1369 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1370 struct dentry *new_dentry)
1372 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1376 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1377 struct inode *new_dir, struct dentry *new_dentry)
1380 err = ll_rename_generic(old_dir, NULL,
1381 old_dentry, &old_dentry->d_name,
1382 new_dir, NULL, new_dentry,
1383 &new_dentry->d_name);
1385 d_move(old_dentry, new_dentry);
1389 const struct inode_operations ll_dir_inode_operations = {
1391 #ifdef HAVE_IOP_ATOMIC_OPEN
1392 .atomic_open = ll_atomic_open,
1394 .lookup = ll_lookup_nd,
1395 .create = ll_create_nd,
1396 /* We need all these non-raw things for NFSD, to not patch it. */
1397 .unlink = ll_unlink,
1400 .symlink = ll_symlink,
1402 .rename = ll_rename,
1403 .setattr = ll_setattr,
1404 .getattr = ll_getattr,
1405 .permission = ll_inode_permission,
1406 .setxattr = ll_setxattr,
1407 .getxattr = ll_getxattr,
1408 .listxattr = ll_listxattr,
1409 .removexattr = ll_removexattr,
1410 #ifdef HAVE_IOP_GET_ACL
1411 .get_acl = ll_get_acl,
1415 const struct inode_operations ll_special_inode_operations = {
1416 .setattr = ll_setattr,
1417 .getattr = ll_getattr,
1418 .permission = ll_inode_permission,
1419 .setxattr = ll_setxattr,
1420 .getxattr = ll_getxattr,
1421 .listxattr = ll_listxattr,
1422 .removexattr = ll_removexattr,
1423 #ifdef HAVE_IOP_GET_ACL
1424 .get_acl = ll_get_acl,