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->mbo_valid & OBD_MD_FLID))) {
85 CERROR("MDS body missing FID\n");
89 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_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->mbo_valid & OBD_MD_FLID))) {
101 CERROR("MDS body missing FID\n");
105 lli->lli_fid = body->mbo_fid1;
106 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
107 CERROR("Can not initialize inode "DFID" without object type: "
109 PFID(&lli->lli_fid), body->mbo_valid);
113 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
114 if (unlikely(inode->i_mode == 0)) {
115 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
126 * Get an inode by inode number(@hash), which is already instantiated by
127 * the intent lookup).
129 struct inode *ll_iget(struct super_block *sb, ino_t hash,
130 struct lustre_md *md)
138 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
140 RETURN(ERR_PTR(-ENOMEM));
142 if (inode->i_state & I_NEW) {
143 rc = ll_read_inode2(inode, md);
144 if (rc == 0 && S_ISREG(inode->i_mode) &&
145 ll_i2info(inode)->lli_clob == NULL) {
146 CDEBUG(D_INODE, "%s: apply lsm %p to inode "DFID"\n",
147 ll_get_fsname(sb, NULL, 0), md->lsm,
148 PFID(ll_inode2fid(inode)));
149 rc = cl_file_inode_init(inode, md);
152 make_bad_inode(inode);
153 unlock_new_inode(inode);
157 unlock_new_inode(inode);
159 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
160 rc = ll_update_inode(inode, md);
161 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
162 PFID(&md->body->mbo_fid1), inode, rc);
164 make_bad_inode(inode);
173 static void ll_invalidate_negative_children(struct inode *dir)
175 struct dentry *dentry, *tmp_subdir;
176 DECLARE_LL_D_HLIST_NODE_PTR(p);
179 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
180 spin_lock(&dentry->d_lock);
181 if (!list_empty(&dentry->d_subdirs)) {
182 struct dentry *child;
184 list_for_each_entry_safe(child, tmp_subdir,
187 if (child->d_inode == NULL)
188 d_lustre_invalidate(child, 1);
191 spin_unlock(&dentry->d_lock);
193 ll_unlock_dcache(dir);
196 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
198 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
201 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
202 void *data, int flag)
204 struct lustre_handle lockh;
209 case LDLM_CB_BLOCKING:
210 ldlm_lock2handle(lock, &lockh);
211 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
213 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
217 case LDLM_CB_CANCELING: {
218 struct inode *inode = ll_inode_from_resource_lock(lock);
219 __u64 bits = lock->l_policy_data.l_inodebits.bits;
221 /* Inode is set to lock->l_resource->lr_lvb_inode
222 * for mdc - bug 24555 */
223 LASSERT(lock->l_ast_data == NULL);
228 /* Invalidate all dentries associated with this inode */
229 LASSERT(ldlm_is_canceling(lock));
231 if (!fid_res_name_eq(ll_inode2fid(inode),
232 &lock->l_resource->lr_name)) {
233 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
234 PFID(ll_inode2fid(inode)), inode);
238 if (bits & MDS_INODELOCK_XATTR) {
239 ll_xattr_cache_destroy(inode);
240 bits &= ~MDS_INODELOCK_XATTR;
243 /* For OPEN locks we differentiate between lock modes
244 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
245 if (bits & MDS_INODELOCK_OPEN)
246 ll_have_md_lock(inode, &bits, lock->l_req_mode);
248 if (bits & MDS_INODELOCK_OPEN) {
251 switch (lock->l_req_mode) {
262 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
266 ll_md_real_close(inode, fmode);
268 bits &= ~MDS_INODELOCK_OPEN;
271 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
272 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
273 ll_have_md_lock(inode, &bits, LCK_MINMODE);
275 if (bits & MDS_INODELOCK_LAYOUT) {
276 struct cl_object_conf conf = {
277 .coc_opc = OBJECT_CONF_INVALIDATE,
281 rc = ll_layout_conf(inode, &conf);
283 CDEBUG(D_INODE, "cannot invalidate layout of "
285 PFID(ll_inode2fid(inode)), rc);
288 if (bits & MDS_INODELOCK_UPDATE) {
289 struct ll_inode_info *lli = ll_i2info(inode);
291 spin_lock(&lli->lli_lock);
292 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
293 spin_unlock(&lli->lli_lock);
296 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
297 struct ll_inode_info *lli = ll_i2info(inode);
299 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
300 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
301 lli, PFID(&lli->lli_pfid));
302 truncate_inode_pages(inode->i_mapping, 0);
304 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
305 struct inode *master_inode = NULL;
308 /* This is slave inode, since all of the child
309 * dentry is connected on the master inode, so
310 * we have to invalidate the negative children
312 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
313 PFID(ll_inode2fid(inode)),
314 PFID(&lli->lli_pfid));
316 hash = cl_fid_build_ino(&lli->lli_pfid,
317 ll_need_32bit_api(ll_i2sbi(inode)));
319 master_inode = ilookup5(inode->i_sb, hash,
320 ll_test_inode_by_fid,
321 (void *)&lli->lli_pfid);
322 if (master_inode != NULL &&
323 !IS_ERR(master_inode)) {
324 ll_invalidate_negative_children(
329 ll_invalidate_negative_children(inode);
333 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
334 inode->i_sb->s_root != NULL &&
335 inode != inode->i_sb->s_root->d_inode)
336 ll_invalidate_aliases(inode);
348 __u32 ll_i2suppgid(struct inode *i)
350 if (in_group_p(i->i_gid))
351 return (__u32)from_kgid(&init_user_ns, i->i_gid);
353 return (__u32) __kgid_val(INVALID_GID);
356 /* Pack the required supplementary groups into the supplied groups array.
357 * If we don't need to use the groups from the target inode(s) then we
358 * instead pack one or more groups from the user's supplementary group
359 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
360 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
363 LASSERT(suppgids != NULL);
365 suppgids[0] = ll_i2suppgid(i1);
368 suppgids[1] = ll_i2suppgid(i2);
374 * try to reuse three types of dentry:
375 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
376 * by concurrent .revalidate).
377 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
378 * be cleared by others calling d_lustre_revalidate).
379 * 3. DISCONNECTED alias.
381 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
383 struct dentry *alias, *discon_alias, *invalid_alias;
384 DECLARE_LL_D_HLIST_NODE_PTR(p);
386 if (ll_d_hlist_empty(&inode->i_dentry))
389 discon_alias = invalid_alias = NULL;
391 ll_lock_dcache(inode);
392 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
393 LASSERT(alias != dentry);
395 spin_lock(&alias->d_lock);
396 if (alias->d_flags & DCACHE_DISCONNECTED)
397 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
398 discon_alias = alias;
399 else if (alias->d_parent == dentry->d_parent &&
400 alias->d_name.hash == dentry->d_name.hash &&
401 alias->d_name.len == dentry->d_name.len &&
402 memcmp(alias->d_name.name, dentry->d_name.name,
403 dentry->d_name.len) == 0)
404 invalid_alias = alias;
405 spin_unlock(&alias->d_lock);
410 alias = invalid_alias ?: discon_alias ?: NULL;
412 spin_lock(&alias->d_lock);
414 spin_unlock(&alias->d_lock);
416 ll_unlock_dcache(inode);
422 * Similar to d_splice_alias(), but lustre treats invalid alias
423 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
425 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
431 new = ll_find_alias(inode, de);
441 "Reuse dentry %p inode %p refc %d flags %#x\n",
442 new, new->d_inode, d_count(new), new->d_flags);
450 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
451 de, de->d_inode, d_count(de), de->d_flags);
455 static int ll_lookup_it_finish(struct ptlrpc_request *request,
456 struct lookup_intent *it,
457 struct inode *parent, struct dentry **de)
459 struct inode *inode = NULL;
464 /* NB 1 request reference will be taken away by ll_intent_lock()
466 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
467 it->d.lustre.it_disposition);
468 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
469 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
473 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
475 /* We used to query real size from OSTs here, but actually
476 this is not needed. For stat() calls size would be updated
477 from subsequent do_revalidate()->ll_inode_revalidate_it() in
479 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
480 Everybody else who needs correct file size would call
481 ll_glimpse_size or some equivalent themselves anyway.
482 Also see bug 7198. */
485 /* Only hash *de if it is unhashed (new dentry).
486 * Atoimc_open may passin hashed dentries for open.
488 if (d_unhashed(*de)) {
489 struct dentry *alias;
491 alias = ll_splice_alias(inode, *de);
493 RETURN(PTR_ERR(alias));
495 } else if (!it_disposition(it, DISP_LOOKUP_NEG) &&
496 !it_disposition(it, DISP_OPEN_CREATE)) {
497 /* With DISP_OPEN_CREATE dentry will
498 instantiated in ll_create_it. */
499 LASSERT((*de)->d_inode == NULL);
500 d_instantiate(*de, inode);
503 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
504 /* we have lookup look - unhide dentry */
505 if (bits & MDS_INODELOCK_LOOKUP)
506 d_lustre_revalidate(*de);
507 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
508 /* If file created on server, don't depend on parent UPDATE
509 * lock to unhide it. It is left hidden and next lookup can
510 * find it in ll_splice_alias.
512 /* Check that parent has UPDATE lock. */
513 struct lookup_intent parent_it = {
515 .d.lustre.it_lock_handle = 0 };
516 struct lu_fid fid = ll_i2info(parent)->lli_fid;
518 /* If it is striped directory, get the real stripe parent */
519 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
520 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
521 ll_i2info(parent)->lli_lsm_md,
523 (*de)->d_name.len, &fid);
528 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
530 d_lustre_revalidate(*de);
531 ll_intent_release(&parent_it);
538 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
539 struct lookup_intent *it, int lookup_flags)
541 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
542 struct dentry *save = dentry, *retval;
543 struct ptlrpc_request *req = NULL;
544 struct md_op_data *op_data;
549 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
550 RETURN(ERR_PTR(-ENAMETOOLONG));
552 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
553 dentry->d_name.len, dentry->d_name.name,
554 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
556 if (d_mountpoint(dentry))
557 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
559 if (it == NULL || it->it_op == IT_GETXATTR)
562 if (it->it_op == IT_GETATTR) {
563 rc = ll_statahead_enter(parent, &dentry, 0);
566 GOTO(out, retval = NULL);
567 GOTO(out, retval = dentry);
571 if (it->it_op & IT_CREAT)
572 opc = LUSTRE_OPC_CREATE;
574 opc = LUSTRE_OPC_ANY;
576 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
577 dentry->d_name.len, lookup_flags, opc,
580 RETURN((void *)op_data);
582 /* enforce umask if acl disabled or MDS doesn't support umask */
583 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
584 it->it_create_mode &= ~current_umask();
586 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
587 &ll_md_blocking_ast, 0);
588 ll_finish_md_op_data(op_data);
590 GOTO(out, retval = ERR_PTR(rc));
592 rc = ll_lookup_it_finish(req, it, parent, &dentry);
594 ll_intent_release(it);
595 GOTO(out, retval = ERR_PTR(rc));
598 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
599 !S_ISREG(dentry->d_inode->i_mode) &&
600 !S_ISDIR(dentry->d_inode->i_mode)) {
601 ll_release_openhandle(dentry, it);
603 ll_lookup_finish_locks(it, dentry);
606 GOTO(out, retval = NULL);
608 GOTO(out, retval = dentry);
611 ptlrpc_req_finished(req);
612 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
613 ll_statahead_mark(parent, dentry);
617 #ifdef HAVE_IOP_ATOMIC_OPEN
618 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
621 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
624 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
625 dentry->d_name.len, dentry->d_name.name,
626 PFID(ll_inode2fid(parent)), parent, flags);
628 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
629 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
632 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
636 de = ll_lookup_it(parent, dentry, itp, 0);
639 ll_intent_release(itp);
645 * For cached negative dentry and new dentry, handle lookup/create/open
648 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
649 struct file *file, unsigned open_flags,
650 umode_t mode, int *opened)
652 struct lookup_intent *it;
654 long long lookup_flags = LOOKUP_OPEN;
658 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
659 "open_flags %x, mode %x opened %d\n",
660 dentry->d_name.len, dentry->d_name.name,
661 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
663 OBD_ALLOC(it, sizeof(*it));
668 if (open_flags & O_CREAT) {
669 it->it_op |= IT_CREAT;
670 lookup_flags |= LOOKUP_CREATE;
672 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
673 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
674 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
676 /* Dentry added to dcache tree in ll_lookup_it */
677 de = ll_lookup_it(dir, dentry, it, lookup_flags);
684 if (it_disposition(it, DISP_OPEN_CREATE)) {
685 /* Dentry instantiated in ll_create_it. */
686 rc = ll_create_it(dir, dentry, mode, it);
688 /* We dget in ll_splice_alias. */
694 *opened |= FILE_CREATED;
696 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
698 if (S_ISFIFO(dentry->d_inode->i_mode)) {
699 /* We cannot call open here as it would
702 if (it_disposition(it, DISP_ENQ_OPEN_REF))
704 (struct ptlrpc_request *)
705 it->d.lustre.it_data);
706 rc = finish_no_open(file, de);
708 file->private_data = it;
709 rc = finish_open(file, dentry, NULL, opened);
710 /* We dget in ll_splice_alias. finish_open takes
711 * care of dget for fd open.
717 rc = finish_no_open(file, de);
722 ll_intent_release(it);
723 OBD_FREE(it, sizeof(*it));
728 #else /* !HAVE_IOP_ATOMIC_OPEN */
729 static struct lookup_intent *
730 ll_convert_intent(struct open_intent *oit, int lookup_flags)
732 struct lookup_intent *it;
736 return ERR_PTR(-ENOMEM);
738 if (lookup_flags & LOOKUP_OPEN) {
740 if (lookup_flags & LOOKUP_CREATE)
741 it->it_op |= IT_CREAT;
742 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
743 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
744 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
746 it->it_op = IT_GETATTR;
752 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
753 struct nameidata *nd)
758 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
759 struct lookup_intent *it;
761 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
762 it = ll_d2d(dentry)->lld_it;
763 ll_d2d(dentry)->lld_it = NULL;
765 if ((nd->flags & LOOKUP_CREATE) &&
766 !(nd->flags & LOOKUP_OPEN))
769 it = ll_convert_intent(&nd->intent.open, nd->flags);
771 RETURN((struct dentry *)it);
774 de = ll_lookup_it(parent, dentry, it, nd->flags);
777 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
778 if (dentry->d_inode &&
779 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
780 if (S_ISFIFO(dentry->d_inode->i_mode)) {
781 // We cannot call open here as it would
784 (struct ptlrpc_request *)
785 it->d.lustre.it_data);
789 nd->intent.open.file->private_data = it;
790 filp = lookup_instantiate_filp(nd,
796 de = (struct dentry *)filp;
799 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
800 // XXX This can only reliably work on assumption
801 // that there are NO hashed negative dentries.
802 ll_d2d(dentry)->lld_it = it;
803 it = NULL; /* Will be freed in ll_create_nd */
804 /* We absolutely depend on ll_create_nd to be
805 * called to not leak this intent and possible
806 * data attached to it */
811 ll_intent_release(it);
812 OBD_FREE(it, sizeof(*it));
815 de = ll_lookup_it(parent, dentry, NULL, 0);
820 #endif /* HAVE_IOP_ATOMIC_OPEN */
822 /* We depend on "mode" being set with the proper file type/umask by now */
823 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
825 struct inode *inode = NULL;
826 struct ptlrpc_request *request = NULL;
827 struct ll_sb_info *sbi = ll_i2sbi(dir);
831 LASSERT(it && it->d.lustre.it_disposition);
833 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
834 request = it->d.lustre.it_data;
835 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
836 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
838 GOTO(out, inode = ERR_PTR(rc));
840 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
842 /* We asked for a lock on the directory, but were granted a
843 * lock on the inode. Since we finally have an inode pointer,
844 * stuff it in the lock. */
845 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
846 PFID(ll_inode2fid(inode)), inode);
847 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
850 ptlrpc_req_finished(request);
855 * By the time this is called, we already have created the directory cache
856 * entry for the new file, but it is so far negative - it has no inode.
858 * We defer creating the OBD object(s) until open, to keep the intent and
859 * non-intent code paths similar, and also because we do not have the MDS
860 * inode number before calling ll_create_node() (which is needed for LOV),
861 * so we would need to do yet another RPC to the MDS to store the LOV EA
862 * data on the MDS. If needed, we would pass the PACKED lmm as data and
863 * lmm_size in datalen (the MDS still has code which will handle that).
865 * If the create succeeds, we fill in the inode information
866 * with d_instantiate().
868 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
869 struct lookup_intent *it)
875 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
876 dentry->d_name.len, dentry->d_name.name,
877 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
879 rc = it_open_error(DISP_OPEN_CREATE, it);
883 inode = ll_create_node(dir, it);
885 RETURN(PTR_ERR(inode));
887 d_instantiate(dentry, inode);
891 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
893 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
897 if (body->mbo_valid & OBD_MD_FLMTIME &&
898 body->mbo_mtime > LTIME_S(inode->i_mtime)) {
899 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
900 "\n", PFID(ll_inode2fid(inode)),
901 LTIME_S(inode->i_mtime), body->mbo_mtime);
902 LTIME_S(inode->i_mtime) = body->mbo_mtime;
905 if (body->mbo_valid & OBD_MD_FLCTIME &&
906 body->mbo_ctime > LTIME_S(inode->i_ctime))
907 LTIME_S(inode->i_ctime) = body->mbo_ctime;
910 static int ll_new_node(struct inode *dir, struct qstr *name,
911 const char *tgt, int mode, int rdev,
912 struct dentry *dchild, __u32 opc)
914 struct ptlrpc_request *request = NULL;
915 struct md_op_data *op_data;
916 struct inode *inode = NULL;
917 struct ll_sb_info *sbi = ll_i2sbi(dir);
922 if (unlikely(tgt != NULL))
923 tgt_len = strlen(tgt) + 1;
925 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
926 name->len, 0, opc, NULL);
928 GOTO(err_exit, err = PTR_ERR(op_data));
930 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
931 from_kuid(&init_user_ns, current_fsuid()),
932 from_kgid(&init_user_ns, current_fsgid()),
933 cfs_curproc_cap_pack(), rdev, &request);
934 ll_finish_md_op_data(op_data);
938 ll_update_times(request, dir);
941 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
945 d_instantiate(dchild, inode);
949 ptlrpc_req_finished(request);
954 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
955 unsigned rdev, struct dentry *dchild)
960 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
961 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
964 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
965 mode &= ~current_umask();
967 switch (mode & S_IFMT) {
969 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
975 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
986 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
991 #ifdef HAVE_IOP_ATOMIC_OPEN
993 * Plain create. Intent create is handled in atomic_open.
995 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
996 umode_t mode, bool want_excl)
1000 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1001 "flags=%u, excl=%d\n", dentry->d_name.len,
1002 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1003 dir, mode, want_excl);
1005 rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1007 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1009 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1010 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1014 #else /* !HAVE_IOP_ATOMIC_OPEN */
1015 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1016 ll_umode_t mode, struct nameidata *nd)
1018 struct ll_dentry_data *lld = ll_d2d(dentry);
1019 struct lookup_intent *it = NULL;
1026 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1030 /* Was there an error? Propagate it! */
1031 if (it->d.lustre.it_status) {
1032 rc = it->d.lustre.it_status;
1036 rc = ll_create_it(dir, dentry, mode, it);
1037 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1040 nd->intent.open.file->private_data = it;
1041 filp = lookup_instantiate_filp(nd, dentry, NULL);
1047 ll_intent_release(it);
1048 OBD_FREE(it, sizeof(*it));
1051 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1055 #endif /* HAVE_IOP_ATOMIC_OPEN */
1057 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1058 const char *tgt, struct dentry *dchild)
1063 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1064 name->len, name->name, PFID(ll_inode2fid(dir)),
1067 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1068 0, dchild, LUSTRE_OPC_SYMLINK);
1071 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1076 static int ll_link_generic(struct inode *src, struct inode *dir,
1077 struct qstr *name, struct dentry *dchild)
1079 struct ll_sb_info *sbi = ll_i2sbi(dir);
1080 struct ptlrpc_request *request = NULL;
1081 struct md_op_data *op_data;
1085 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID
1086 "(%p), target=%.*s\n", PFID(ll_inode2fid(src)), src,
1087 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1089 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1090 0, LUSTRE_OPC_ANY, NULL);
1091 if (IS_ERR(op_data))
1092 RETURN(PTR_ERR(op_data));
1094 err = md_link(sbi->ll_md_exp, op_data, &request);
1095 ll_finish_md_op_data(op_data);
1099 ll_update_times(request, dir);
1100 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1103 ptlrpc_req_finished(request);
1107 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1108 int mode, struct dentry *dchild)
1114 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1115 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1117 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1118 mode &= ~current_umask();
1119 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1120 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1123 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1128 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1129 struct dentry *dchild, struct qstr *name)
1131 struct ptlrpc_request *request = NULL;
1132 struct md_op_data *op_data;
1136 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1137 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1139 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1142 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1143 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1144 if (IS_ERR(op_data))
1145 RETURN(PTR_ERR(op_data));
1147 if (dchild != NULL && dchild->d_inode != NULL)
1148 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1149 op_data->op_fid2 = op_data->op_fid3;
1150 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1151 ll_finish_md_op_data(op_data);
1153 ll_update_times(request, dir);
1154 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1157 ptlrpc_req_finished(request);
1164 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1166 struct ptlrpc_request *request = NULL;
1167 struct md_op_data *op_data;
1171 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1172 namelen, name, PFID(ll_inode2fid(dir)), dir);
1174 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1175 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1176 if (IS_ERR(op_data))
1177 RETURN(PTR_ERR(op_data));
1178 op_data->op_cli_flags |= CLI_RM_ENTRY;
1179 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1180 ll_finish_md_op_data(op_data);
1182 ll_update_times(request, dir);
1183 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1186 ptlrpc_req_finished(request);
1190 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1192 struct mdt_body *body;
1193 struct lov_mds_md *eadata;
1194 struct lov_stripe_md *lsm = NULL;
1195 struct obd_trans_info oti = { 0 };
1197 struct obd_capa *oc = NULL;
1201 /* req is swabbed so this is safe */
1202 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1203 if (!(body->mbo_valid & OBD_MD_FLEASIZE))
1206 if (body->mbo_eadatasize == 0) {
1207 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1208 GOTO(out, rc = -EPROTO);
1211 /* The MDS sent back the EA because we unlinked the last reference
1212 * to this file. Use this EA to unlink the objects on the OST.
1213 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1214 * check it is complete and sensible. */
1215 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1216 body->mbo_eadatasize);
1217 LASSERT(eadata != NULL);
1219 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->mbo_eadatasize);
1221 CERROR("obd_unpackmd: %d\n", rc);
1224 LASSERT(rc >= sizeof(*lsm));
1228 GOTO(out_free_memmd, rc = -ENOMEM);
1230 oa->o_oi = lsm->lsm_oi;
1231 oa->o_mode = body->mbo_mode & S_IFMT;
1232 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1234 if (body->mbo_valid & OBD_MD_FLCOOKIE) {
1235 oa->o_valid |= OBD_MD_FLCOOKIE;
1236 oti.oti_logcookies =
1237 req_capsule_server_sized_get(&request->rq_pill,
1239 sizeof(struct llog_cookie) *
1240 lsm->lsm_stripe_count);
1241 if (oti.oti_logcookies == NULL) {
1242 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1243 body->mbo_valid &= ~OBD_MD_FLCOOKIE;
1247 if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
1248 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1250 GOTO(out_free_memmd, rc);
1253 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1254 ll_i2mdexp(dir), oc);
1257 CERROR("obd destroy objid "DOSTID" error %d\n",
1258 POSTID(&lsm->lsm_oi), rc);
1260 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1266 /* ll_unlink_generic() doesn't update the inode with the new link count.
1267 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1268 * is any lock existing. They will recycle dentries and inodes based upon locks
1270 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1271 struct dentry *dchild, struct qstr *name)
1273 struct ptlrpc_request *request = NULL;
1274 struct md_op_data *op_data;
1277 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1278 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1281 * XXX: unlink bind mountpoint maybe call to here,
1282 * just check it as vfs_unlink does.
1284 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1287 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1288 name->len, 0, LUSTRE_OPC_ANY, NULL);
1289 if (IS_ERR(op_data))
1290 RETURN(PTR_ERR(op_data));
1292 if (dchild != NULL && dchild->d_inode != NULL)
1293 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1295 op_data->op_fid2 = op_data->op_fid3;
1296 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1297 ll_finish_md_op_data(op_data);
1301 ll_update_times(request, dir);
1302 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1304 rc = ll_objects_destroy(request, dir);
1306 ptlrpc_req_finished(request);
1310 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1311 struct dentry *src_dchild, struct qstr *src_name,
1312 struct inode *tgt, struct dentry *tgt_dparent,
1313 struct dentry *tgt_dchild, struct qstr *tgt_name)
1315 struct ptlrpc_request *request = NULL;
1316 struct ll_sb_info *sbi = ll_i2sbi(src);
1317 struct md_op_data *op_data;
1320 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1321 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1322 src_name->len, src_name->name,
1323 PFID(ll_inode2fid(src)), src, tgt_name->len,
1324 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1326 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1327 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1330 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1331 LUSTRE_OPC_ANY, NULL);
1332 if (IS_ERR(op_data))
1333 RETURN(PTR_ERR(op_data));
1335 if (src_dchild != NULL && src_dchild->d_inode != NULL)
1336 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1337 if (tgt_dchild != NULL && tgt_dchild->d_inode != NULL)
1338 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1340 err = md_rename(sbi->ll_md_exp, op_data,
1341 src_name->name, src_name->len,
1342 tgt_name->name, tgt_name->len, &request);
1343 ll_finish_md_op_data(op_data);
1345 ll_update_times(request, src);
1346 ll_update_times(request, tgt);
1347 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1348 err = ll_objects_destroy(request, src);
1351 ptlrpc_req_finished(request);
1356 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1359 return ll_mknod_generic(dir, &dchild->d_name, mode,
1360 old_encode_dev(rdev), dchild);
1363 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1365 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1368 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1370 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1373 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1375 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1378 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1379 const char *oldname)
1381 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1384 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1385 struct dentry *new_dentry)
1387 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1391 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1392 struct inode *new_dir, struct dentry *new_dentry)
1395 err = ll_rename_generic(old_dir, NULL,
1396 old_dentry, &old_dentry->d_name,
1397 new_dir, NULL, new_dentry,
1398 &new_dentry->d_name);
1400 d_move(old_dentry, new_dentry);
1404 const struct inode_operations ll_dir_inode_operations = {
1406 #ifdef HAVE_IOP_ATOMIC_OPEN
1407 .atomic_open = ll_atomic_open,
1409 .lookup = ll_lookup_nd,
1410 .create = ll_create_nd,
1411 /* We need all these non-raw things for NFSD, to not patch it. */
1412 .unlink = ll_unlink,
1415 .symlink = ll_symlink,
1417 .rename = ll_rename,
1418 .setattr = ll_setattr,
1419 .getattr = ll_getattr,
1420 .permission = ll_inode_permission,
1421 .setxattr = ll_setxattr,
1422 .getxattr = ll_getxattr,
1423 .listxattr = ll_listxattr,
1424 .removexattr = ll_removexattr,
1425 #ifdef HAVE_IOP_GET_ACL
1426 .get_acl = ll_get_acl,
1430 const struct inode_operations ll_special_inode_operations = {
1431 .setattr = ll_setattr,
1432 .getattr = ll_getattr,
1433 .permission = ll_inode_permission,
1434 .setxattr = ll_setxattr,
1435 .getxattr = ll_getxattr,
1436 .listxattr = ll_listxattr,
1437 .removexattr = ll_removexattr,
1438 #ifdef HAVE_IOP_GET_ACL
1439 .get_acl = ll_get_acl,