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 (already instantiated by the intent lookup).
126 * Returns inode or NULL
128 struct inode *ll_iget(struct super_block *sb, ino_t hash,
129 struct lustre_md *md)
135 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
138 if (inode->i_state & I_NEW) {
141 ll_read_inode2(inode, md);
142 if (S_ISREG(inode->i_mode) &&
143 ll_i2info(inode)->lli_clob == NULL) {
145 "%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);
157 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
158 ll_update_inode(inode, md);
159 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p)\n",
160 PFID(&md->body->fid1), inode);
165 static void ll_invalidate_negative_children(struct inode *dir)
167 struct dentry *dentry, *tmp_subdir;
168 DECLARE_LL_D_HLIST_NODE_PTR(p);
171 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
172 spin_lock(&dentry->d_lock);
173 if (!list_empty(&dentry->d_subdirs)) {
174 struct dentry *child;
176 list_for_each_entry_safe(child, tmp_subdir,
179 if (child->d_inode == NULL)
180 d_lustre_invalidate(child, 1);
183 spin_unlock(&dentry->d_lock);
185 ll_unlock_dcache(dir);
188 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
190 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
193 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
194 void *data, int flag)
196 struct lustre_handle lockh;
201 case LDLM_CB_BLOCKING:
202 ldlm_lock2handle(lock, &lockh);
203 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
205 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
209 case LDLM_CB_CANCELING: {
210 struct inode *inode = ll_inode_from_resource_lock(lock);
211 __u64 bits = lock->l_policy_data.l_inodebits.bits;
213 /* Inode is set to lock->l_resource->lr_lvb_inode
214 * for mdc - bug 24555 */
215 LASSERT(lock->l_ast_data == NULL);
220 /* Invalidate all dentries associated with this inode */
221 LASSERT(ldlm_is_canceling(lock));
223 if (!fid_res_name_eq(ll_inode2fid(inode),
224 &lock->l_resource->lr_name)) {
225 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
226 PFID(ll_inode2fid(inode)), inode);
230 if (bits & MDS_INODELOCK_XATTR) {
231 ll_xattr_cache_destroy(inode);
232 bits &= ~MDS_INODELOCK_XATTR;
235 /* For OPEN locks we differentiate between lock modes
236 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
237 if (bits & MDS_INODELOCK_OPEN)
238 ll_have_md_lock(inode, &bits, lock->l_req_mode);
240 if (bits & MDS_INODELOCK_OPEN) {
243 switch (lock->l_req_mode) {
254 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
258 ll_md_real_close(inode, fmode);
260 bits &= ~MDS_INODELOCK_OPEN;
263 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
264 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
265 ll_have_md_lock(inode, &bits, LCK_MINMODE);
267 if (bits & MDS_INODELOCK_LAYOUT) {
268 struct cl_object_conf conf = {
269 .coc_opc = OBJECT_CONF_INVALIDATE,
273 rc = ll_layout_conf(inode, &conf);
275 CDEBUG(D_INODE, "cannot invalidate layout of "
277 PFID(ll_inode2fid(inode)), rc);
280 if (bits & MDS_INODELOCK_UPDATE) {
281 struct ll_inode_info *lli = ll_i2info(inode);
283 spin_lock(&lli->lli_lock);
284 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
285 spin_unlock(&lli->lli_lock);
288 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
289 struct ll_inode_info *lli = ll_i2info(inode);
291 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
292 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
293 lli, PFID(&lli->lli_pfid));
294 truncate_inode_pages(inode->i_mapping, 0);
296 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
297 struct inode *master_inode = NULL;
300 /* This is slave inode, since all of the child
301 * dentry is connected on the master inode, so
302 * we have to invalidate the negative children
304 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
305 PFID(ll_inode2fid(inode)),
306 PFID(&lli->lli_pfid));
308 hash = cl_fid_build_ino(&lli->lli_pfid,
309 ll_need_32bit_api(ll_i2sbi(inode)));
311 master_inode = ilookup5(inode->i_sb, hash,
312 ll_test_inode_by_fid,
313 (void *)&lli->lli_pfid);
314 if (master_inode != NULL &&
315 !IS_ERR(master_inode)) {
316 ll_invalidate_negative_children(
321 ll_invalidate_negative_children(inode);
325 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
326 inode->i_sb->s_root != NULL &&
327 inode != inode->i_sb->s_root->d_inode)
328 ll_invalidate_aliases(inode);
340 __u32 ll_i2suppgid(struct inode *i)
342 if (in_group_p(i->i_gid))
343 return (__u32)from_kgid(&init_user_ns, i->i_gid);
345 return (__u32) __kgid_val(INVALID_GID);
348 /* Pack the required supplementary groups into the supplied groups array.
349 * If we don't need to use the groups from the target inode(s) then we
350 * instead pack one or more groups from the user's supplementary group
351 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
352 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
355 LASSERT(suppgids != NULL);
357 suppgids[0] = ll_i2suppgid(i1);
360 suppgids[1] = ll_i2suppgid(i2);
366 * try to reuse three types of dentry:
367 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
368 * by concurrent .revalidate).
369 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
370 * be cleared by others calling d_lustre_revalidate).
371 * 3. DISCONNECTED alias.
373 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
375 struct dentry *alias, *discon_alias, *invalid_alias;
376 DECLARE_LL_D_HLIST_NODE_PTR(p);
378 if (ll_d_hlist_empty(&inode->i_dentry))
381 discon_alias = invalid_alias = NULL;
383 ll_lock_dcache(inode);
384 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
385 LASSERT(alias != dentry);
387 spin_lock(&alias->d_lock);
388 if (alias->d_flags & DCACHE_DISCONNECTED)
389 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
390 discon_alias = alias;
391 else if (alias->d_parent == dentry->d_parent &&
392 alias->d_name.hash == dentry->d_name.hash &&
393 alias->d_name.len == dentry->d_name.len &&
394 memcmp(alias->d_name.name, dentry->d_name.name,
395 dentry->d_name.len) == 0)
396 invalid_alias = alias;
397 spin_unlock(&alias->d_lock);
402 alias = invalid_alias ?: discon_alias ?: NULL;
404 spin_lock(&alias->d_lock);
406 spin_unlock(&alias->d_lock);
408 ll_unlock_dcache(inode);
414 * Similar to d_splice_alias(), but lustre treats invalid alias
415 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
417 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
423 new = ll_find_alias(inode, de);
433 "Reuse dentry %p inode %p refc %d flags %#x\n",
434 new, new->d_inode, d_count(new), new->d_flags);
442 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
443 de, de->d_inode, d_count(de), de->d_flags);
447 static int ll_lookup_it_finish(struct ptlrpc_request *request,
448 struct lookup_intent *it,
449 struct inode *parent, struct dentry **de)
451 struct inode *inode = NULL;
456 /* NB 1 request reference will be taken away by ll_intent_lock()
458 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
459 it->d.lustre.it_disposition);
460 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
461 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
465 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
467 /* We used to query real size from OSTs here, but actually
468 this is not needed. For stat() calls size would be updated
469 from subsequent do_revalidate()->ll_inode_revalidate_it() in
471 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
472 Everybody else who needs correct file size would call
473 ll_glimpse_size or some equivalent themselves anyway.
474 Also see bug 7198. */
477 /* Only hash *de if it is unhashed (new dentry).
478 * Atoimc_open may passin hashed dentries for open.
480 if (d_unhashed(*de)) {
481 struct dentry *alias;
483 alias = ll_splice_alias(inode, *de);
485 RETURN(PTR_ERR(alias));
487 } else if (!it_disposition(it, DISP_LOOKUP_NEG) &&
488 !it_disposition(it, DISP_OPEN_CREATE)) {
489 /* With DISP_OPEN_CREATE dentry will
490 instantiated in ll_create_it. */
491 LASSERT((*de)->d_inode == NULL);
492 d_instantiate(*de, inode);
495 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
496 /* we have lookup look - unhide dentry */
497 if (bits & MDS_INODELOCK_LOOKUP)
498 d_lustre_revalidate(*de);
499 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
500 /* If file created on server, don't depend on parent UPDATE
501 * lock to unhide it. It is left hidden and next lookup can
502 * find it in ll_splice_alias.
504 /* Check that parent has UPDATE lock. */
505 struct lookup_intent parent_it = {
507 .d.lustre.it_lock_handle = 0 };
509 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
510 &ll_i2info(parent)->lli_fid, NULL)) {
511 d_lustre_revalidate(*de);
512 ll_intent_release(&parent_it);
519 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
520 struct lookup_intent *it, int lookup_flags)
522 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
523 struct dentry *save = dentry, *retval;
524 struct ptlrpc_request *req = NULL;
525 struct md_op_data *op_data;
530 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
531 RETURN(ERR_PTR(-ENAMETOOLONG));
533 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
534 dentry->d_name.len, dentry->d_name.name,
535 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
537 if (d_mountpoint(dentry))
538 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
540 if (it == NULL || it->it_op == IT_GETXATTR)
543 if (it->it_op == IT_GETATTR) {
544 rc = ll_statahead_enter(parent, &dentry, 0);
547 GOTO(out, retval = NULL);
548 GOTO(out, retval = dentry);
552 if (it->it_op & IT_CREAT)
553 opc = LUSTRE_OPC_CREATE;
555 opc = LUSTRE_OPC_ANY;
557 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
558 dentry->d_name.len, lookup_flags, opc,
561 RETURN((void *)op_data);
563 /* enforce umask if acl disabled or MDS doesn't support umask */
564 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
565 it->it_create_mode &= ~current_umask();
567 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
568 lookup_flags, &req, ll_md_blocking_ast, 0);
569 ll_finish_md_op_data(op_data);
571 GOTO(out, retval = ERR_PTR(rc));
573 rc = ll_lookup_it_finish(req, it, parent, &dentry);
575 ll_intent_release(it);
576 GOTO(out, retval = ERR_PTR(rc));
579 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
580 !S_ISREG(dentry->d_inode->i_mode) &&
581 !S_ISDIR(dentry->d_inode->i_mode)) {
582 ll_release_openhandle(dentry, it);
584 ll_lookup_finish_locks(it, dentry);
587 GOTO(out, retval = NULL);
589 GOTO(out, retval = dentry);
592 ptlrpc_req_finished(req);
593 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
594 ll_statahead_mark(parent, dentry);
598 #ifdef HAVE_IOP_ATOMIC_OPEN
599 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
602 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
605 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
606 dentry->d_name.len, dentry->d_name.name,
607 PFID(ll_inode2fid(parent)), parent, flags);
609 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
610 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
613 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
617 de = ll_lookup_it(parent, dentry, itp, 0);
620 ll_intent_release(itp);
626 * For cached negative dentry and new dentry, handle lookup/create/open
629 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
630 struct file *file, unsigned open_flags,
631 umode_t mode, int *opened)
633 struct lookup_intent *it;
635 long long lookup_flags = LOOKUP_OPEN;
639 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
640 "open_flags %x, mode %x opened %d\n",
641 dentry->d_name.len, dentry->d_name.name,
642 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
644 OBD_ALLOC(it, sizeof(*it));
649 if (open_flags & O_CREAT) {
650 it->it_op |= IT_CREAT;
651 lookup_flags |= LOOKUP_CREATE;
653 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
654 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
656 /* Dentry added to dcache tree in ll_lookup_it */
657 de = ll_lookup_it(dir, dentry, it, lookup_flags);
664 if (it_disposition(it, DISP_OPEN_CREATE)) {
665 /* Dentry instantiated in ll_create_it. */
666 rc = ll_create_it(dir, dentry, mode, it);
668 /* We dget in ll_splice_alias. */
674 *opened |= FILE_CREATED;
676 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
678 if (S_ISFIFO(dentry->d_inode->i_mode)) {
679 /* We cannot call open here as it would
682 if (it_disposition(it, DISP_ENQ_OPEN_REF))
684 (struct ptlrpc_request *)
685 it->d.lustre.it_data);
686 rc = finish_no_open(file, de);
688 file->private_data = it;
689 rc = finish_open(file, dentry, NULL, opened);
690 /* We dget in ll_splice_alias. finish_open takes
691 * care of dget for fd open.
697 rc = finish_no_open(file, de);
702 ll_intent_release(it);
703 OBD_FREE(it, sizeof(*it));
708 #else /* !HAVE_IOP_ATOMIC_OPEN */
709 static struct lookup_intent *
710 ll_convert_intent(struct open_intent *oit, int lookup_flags)
712 struct lookup_intent *it;
714 OBD_ALLOC(it, sizeof(*it));
716 return ERR_PTR(-ENOMEM);
718 if (lookup_flags & LOOKUP_OPEN) {
720 if (lookup_flags & LOOKUP_CREATE)
721 it->it_op |= IT_CREAT;
722 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
723 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
725 it->it_op = IT_GETATTR;
731 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
732 struct nameidata *nd)
737 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
738 struct lookup_intent *it;
740 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
741 it = ll_d2d(dentry)->lld_it;
742 ll_d2d(dentry)->lld_it = NULL;
744 if ((nd->flags & LOOKUP_CREATE) &&
745 !(nd->flags & LOOKUP_OPEN))
748 it = ll_convert_intent(&nd->intent.open, nd->flags);
750 RETURN((struct dentry *)it);
753 de = ll_lookup_it(parent, dentry, it, nd->flags);
756 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
757 if (dentry->d_inode &&
758 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
759 if (S_ISFIFO(dentry->d_inode->i_mode)) {
760 // We cannot call open here as it would
763 (struct ptlrpc_request *)
764 it->d.lustre.it_data);
768 nd->intent.open.file->private_data = it;
769 filp = lookup_instantiate_filp(nd,
775 de = (struct dentry *)filp;
778 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
779 // XXX This can only reliably work on assumption
780 // that there are NO hashed negative dentries.
781 ll_d2d(dentry)->lld_it = it;
782 it = NULL; /* Will be freed in ll_create_nd */
783 /* We absolutely depend on ll_create_nd to be
784 * called to not leak this intent and possible
785 * data attached to it */
790 ll_intent_release(it);
791 OBD_FREE(it, sizeof(*it));
794 de = ll_lookup_it(parent, dentry, NULL, 0);
799 #endif /* HAVE_IOP_ATOMIC_OPEN */
801 /* We depend on "mode" being set with the proper file type/umask by now */
802 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
804 struct inode *inode = NULL;
805 struct ptlrpc_request *request = NULL;
806 struct ll_sb_info *sbi = ll_i2sbi(dir);
810 LASSERT(it && it->d.lustre.it_disposition);
812 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
813 request = it->d.lustre.it_data;
814 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
815 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
817 GOTO(out, inode = ERR_PTR(rc));
819 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
821 /* We asked for a lock on the directory, but were granted a
822 * lock on the inode. Since we finally have an inode pointer,
823 * stuff it in the lock. */
824 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
825 PFID(ll_inode2fid(inode)), inode);
826 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
829 ptlrpc_req_finished(request);
834 * By the time this is called, we already have created the directory cache
835 * entry for the new file, but it is so far negative - it has no inode.
837 * We defer creating the OBD object(s) until open, to keep the intent and
838 * non-intent code paths similar, and also because we do not have the MDS
839 * inode number before calling ll_create_node() (which is needed for LOV),
840 * so we would need to do yet another RPC to the MDS to store the LOV EA
841 * data on the MDS. If needed, we would pass the PACKED lmm as data and
842 * lmm_size in datalen (the MDS still has code which will handle that).
844 * If the create succeeds, we fill in the inode information
845 * with d_instantiate().
847 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
848 struct lookup_intent *it)
854 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
855 dentry->d_name.len, dentry->d_name.name,
856 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
858 rc = it_open_error(DISP_OPEN_CREATE, it);
862 inode = ll_create_node(dir, it);
864 RETURN(PTR_ERR(inode));
866 d_instantiate(dentry, inode);
870 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
872 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
876 if (body->valid & OBD_MD_FLMTIME &&
877 body->mtime > LTIME_S(inode->i_mtime)) {
878 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
879 "\n", PFID(ll_inode2fid(inode)),
880 LTIME_S(inode->i_mtime), body->mtime);
881 LTIME_S(inode->i_mtime) = body->mtime;
883 if (body->valid & OBD_MD_FLCTIME &&
884 body->ctime > LTIME_S(inode->i_ctime))
885 LTIME_S(inode->i_ctime) = body->ctime;
888 static int ll_new_node(struct inode *dir, struct qstr *name,
889 const char *tgt, int mode, int rdev,
890 struct dentry *dchild, __u32 opc)
892 struct ptlrpc_request *request = NULL;
893 struct md_op_data *op_data;
894 struct inode *inode = NULL;
895 struct ll_sb_info *sbi = ll_i2sbi(dir);
900 if (unlikely(tgt != NULL))
901 tgt_len = strlen(tgt) + 1;
903 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
904 name->len, 0, opc, NULL);
906 GOTO(err_exit, err = PTR_ERR(op_data));
908 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
909 from_kuid(&init_user_ns, current_fsuid()),
910 from_kgid(&init_user_ns, current_fsgid()),
911 cfs_curproc_cap_pack(), rdev, &request);
912 ll_finish_md_op_data(op_data);
916 ll_update_times(request, dir);
919 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
923 d_instantiate(dchild, inode);
927 ptlrpc_req_finished(request);
932 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
933 unsigned rdev, struct dentry *dchild)
938 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
939 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
942 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
943 mode &= ~current_umask();
945 switch (mode & S_IFMT) {
947 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
953 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
964 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
969 #ifdef HAVE_IOP_ATOMIC_OPEN
971 * Plain create. Intent create is handled in atomic_open.
973 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
974 umode_t mode, bool want_excl)
978 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
979 "flags=%u, excl=%d\n", dentry->d_name.len,
980 dentry->d_name.name, PFID(ll_inode2fid(dir)),
981 dir, mode, want_excl);
983 rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
985 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
987 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
988 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
992 #else /* !HAVE_IOP_ATOMIC_OPEN */
993 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
994 ll_umode_t mode, struct nameidata *nd)
996 struct ll_dentry_data *lld = ll_d2d(dentry);
997 struct lookup_intent *it = NULL;
1004 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1008 /* Was there an error? Propagate it! */
1009 if (it->d.lustre.it_status) {
1010 rc = it->d.lustre.it_status;
1014 rc = ll_create_it(dir, dentry, mode, it);
1015 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1018 nd->intent.open.file->private_data = it;
1019 filp = lookup_instantiate_filp(nd, dentry, NULL);
1025 ll_intent_release(it);
1026 OBD_FREE(it, sizeof(*it));
1029 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1033 #endif /* HAVE_IOP_ATOMIC_OPEN */
1035 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1036 const char *tgt, struct dentry *dchild)
1041 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1042 name->len, name->name, PFID(ll_inode2fid(dir)),
1045 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1046 0, dchild, LUSTRE_OPC_SYMLINK);
1049 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1054 static int ll_link_generic(struct inode *src, struct inode *dir,
1055 struct qstr *name, struct dentry *dchild)
1057 struct ll_sb_info *sbi = ll_i2sbi(dir);
1058 struct ptlrpc_request *request = NULL;
1059 struct md_op_data *op_data;
1063 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID
1064 "(%p), target=%.*s\n", PFID(ll_inode2fid(src)), src,
1065 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1067 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1068 0, LUSTRE_OPC_ANY, NULL);
1069 if (IS_ERR(op_data))
1070 RETURN(PTR_ERR(op_data));
1072 err = md_link(sbi->ll_md_exp, op_data, &request);
1073 ll_finish_md_op_data(op_data);
1077 ll_update_times(request, dir);
1078 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1081 ptlrpc_req_finished(request);
1085 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1086 int mode, struct dentry *dchild)
1092 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1093 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1095 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1096 mode &= ~current_umask();
1097 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1098 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1101 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1106 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1107 struct dentry *dchild, struct qstr *name)
1109 struct ptlrpc_request *request = NULL;
1110 struct md_op_data *op_data;
1114 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1115 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1117 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1120 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1121 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1122 if (IS_ERR(op_data))
1123 RETURN(PTR_ERR(op_data));
1125 if (dchild != NULL && dchild->d_inode != NULL)
1126 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1127 op_data->op_fid2 = op_data->op_fid3;
1128 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1129 ll_finish_md_op_data(op_data);
1131 ll_update_times(request, dir);
1132 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1135 ptlrpc_req_finished(request);
1142 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1144 struct ptlrpc_request *request = NULL;
1145 struct md_op_data *op_data;
1149 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1150 namelen, name, PFID(ll_inode2fid(dir)), dir);
1152 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1153 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1154 if (IS_ERR(op_data))
1155 RETURN(PTR_ERR(op_data));
1156 op_data->op_cli_flags |= CLI_RM_ENTRY;
1157 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1158 ll_finish_md_op_data(op_data);
1160 ll_update_times(request, dir);
1161 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1164 ptlrpc_req_finished(request);
1168 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1170 struct mdt_body *body;
1171 struct lov_mds_md *eadata;
1172 struct lov_stripe_md *lsm = NULL;
1173 struct obd_trans_info oti = { 0 };
1175 struct obd_capa *oc = NULL;
1179 /* req is swabbed so this is safe */
1180 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1181 if (!(body->valid & OBD_MD_FLEASIZE))
1184 if (body->eadatasize == 0) {
1185 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1186 GOTO(out, rc = -EPROTO);
1189 /* The MDS sent back the EA because we unlinked the last reference
1190 * to this file. Use this EA to unlink the objects on the OST.
1191 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1192 * check it is complete and sensible. */
1193 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1195 LASSERT(eadata != NULL);
1197 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1199 CERROR("obd_unpackmd: %d\n", rc);
1202 LASSERT(rc >= sizeof(*lsm));
1206 GOTO(out_free_memmd, rc = -ENOMEM);
1208 oa->o_oi = lsm->lsm_oi;
1209 oa->o_mode = body->mode & S_IFMT;
1210 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1212 if (body->valid & OBD_MD_FLCOOKIE) {
1213 oa->o_valid |= OBD_MD_FLCOOKIE;
1214 oti.oti_logcookies =
1215 req_capsule_server_sized_get(&request->rq_pill,
1217 sizeof(struct llog_cookie) *
1218 lsm->lsm_stripe_count);
1219 if (oti.oti_logcookies == NULL) {
1220 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1221 body->valid &= ~OBD_MD_FLCOOKIE;
1225 if (body->valid & OBD_MD_FLOSSCAPA) {
1226 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1228 GOTO(out_free_memmd, rc);
1231 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1232 ll_i2mdexp(dir), oc);
1235 CERROR("obd destroy objid "DOSTID" error %d\n",
1236 POSTID(&lsm->lsm_oi), rc);
1238 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1244 /* ll_unlink_generic() doesn't update the inode with the new link count.
1245 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1246 * is any lock existing. They will recycle dentries and inodes based upon locks
1248 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1249 struct dentry *dchild, struct qstr *name)
1251 struct ptlrpc_request *request = NULL;
1252 struct md_op_data *op_data;
1255 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1256 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1259 * XXX: unlink bind mountpoint maybe call to here,
1260 * just check it as vfs_unlink does.
1262 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1265 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1266 name->len, 0, LUSTRE_OPC_ANY, NULL);
1267 if (IS_ERR(op_data))
1268 RETURN(PTR_ERR(op_data));
1270 if (dchild != NULL && dchild->d_inode != NULL)
1271 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1273 op_data->op_fid2 = op_data->op_fid3;
1274 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1275 ll_finish_md_op_data(op_data);
1279 ll_update_times(request, dir);
1280 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1282 rc = ll_objects_destroy(request, dir);
1284 ptlrpc_req_finished(request);
1288 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1289 struct dentry *src_dchild, struct qstr *src_name,
1290 struct inode *tgt, struct dentry *tgt_dparent,
1291 struct dentry *tgt_dchild, struct qstr *tgt_name)
1293 struct ptlrpc_request *request = NULL;
1294 struct ll_sb_info *sbi = ll_i2sbi(src);
1295 struct md_op_data *op_data;
1298 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1299 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1300 src_name->len, src_name->name,
1301 PFID(ll_inode2fid(src)), src, tgt_name->len,
1302 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1304 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1305 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1308 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1309 LUSTRE_OPC_ANY, NULL);
1310 if (IS_ERR(op_data))
1311 RETURN(PTR_ERR(op_data));
1313 if (src_dchild != NULL && src_dchild->d_inode != NULL)
1314 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1315 if (tgt_dchild != NULL && tgt_dchild->d_inode != NULL)
1316 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1318 err = md_rename(sbi->ll_md_exp, op_data,
1319 src_name->name, src_name->len,
1320 tgt_name->name, tgt_name->len, &request);
1321 ll_finish_md_op_data(op_data);
1323 ll_update_times(request, src);
1324 ll_update_times(request, tgt);
1325 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1326 err = ll_objects_destroy(request, src);
1329 ptlrpc_req_finished(request);
1334 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1337 return ll_mknod_generic(dir, &dchild->d_name, mode,
1338 old_encode_dev(rdev), dchild);
1341 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1343 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1346 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1348 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1351 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1353 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1356 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1357 const char *oldname)
1359 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1362 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1363 struct dentry *new_dentry)
1365 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1369 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1370 struct inode *new_dir, struct dentry *new_dentry)
1373 err = ll_rename_generic(old_dir, NULL,
1374 old_dentry, &old_dentry->d_name,
1375 new_dir, NULL, new_dentry,
1376 &new_dentry->d_name);
1378 d_move(old_dentry, new_dentry);
1382 const struct inode_operations ll_dir_inode_operations = {
1384 #ifdef HAVE_IOP_ATOMIC_OPEN
1385 .atomic_open = ll_atomic_open,
1387 .lookup = ll_lookup_nd,
1388 .create = ll_create_nd,
1389 /* We need all these non-raw things for NFSD, to not patch it. */
1390 .unlink = ll_unlink,
1393 .symlink = ll_symlink,
1395 .rename = ll_rename,
1396 .setattr = ll_setattr,
1397 .getattr = ll_getattr,
1398 .permission = ll_inode_permission,
1399 .setxattr = ll_setxattr,
1400 .getxattr = ll_getxattr,
1401 .listxattr = ll_listxattr,
1402 .removexattr = ll_removexattr,
1403 #ifdef HAVE_IOP_GET_ACL
1404 .get_acl = ll_get_acl,
1408 const struct inode_operations ll_special_inode_operations = {
1409 .setattr = ll_setattr,
1410 .getattr = ll_getattr,
1411 .permission = ll_inode_permission,
1412 .setxattr = ll_setxattr,
1413 .getxattr = ll_getxattr,
1414 .listxattr = ll_listxattr,
1415 .removexattr = ll_removexattr,
1416 #ifdef HAVE_IOP_GET_ACL
1417 .get_acl = ll_get_acl,