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 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
82 ldlm_lock_decref(lockh, mode);
88 /* called from iget5_locked->find_inode() under inode_lock spinlock */
89 static int ll_test_inode(struct inode *inode, void *opaque)
91 struct ll_inode_info *lli = ll_i2info(inode);
92 struct lustre_md *md = opaque;
94 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
95 CERROR("MDS body missing FID\n");
99 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
105 static int ll_set_inode(struct inode *inode, void *opaque)
107 struct ll_inode_info *lli = ll_i2info(inode);
108 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
110 if (unlikely(!(body->valid & OBD_MD_FLID))) {
111 CERROR("MDS body missing FID\n");
115 lli->lli_fid = body->fid1;
116 if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
117 CERROR("Can not initialize inode "DFID" without object type: "
118 "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
122 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
123 if (unlikely(inode->i_mode == 0)) {
124 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
135 * Get an inode by inode number (already instantiated by the intent lookup).
136 * Returns inode or NULL
138 struct inode *ll_iget(struct super_block *sb, ino_t hash,
139 struct lustre_md *md)
145 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
148 if (inode->i_state & I_NEW) {
151 ll_read_inode2(inode, md);
152 if (S_ISREG(inode->i_mode) &&
153 ll_i2info(inode)->lli_clob == NULL) {
155 "%s: apply lsm %p to inode "DFID".\n",
156 ll_get_fsname(sb, NULL, 0), md->lsm,
157 PFID(ll_inode2fid(inode)));
158 rc = cl_file_inode_init(inode, md);
161 make_bad_inode(inode);
162 unlock_new_inode(inode);
166 unlock_new_inode(inode);
167 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
168 ll_update_inode(inode, md);
169 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
170 inode, PFID(&md->body->fid1));
175 static void ll_invalidate_negative_children(struct inode *dir)
177 struct dentry *dentry, *tmp_subdir;
178 DECLARE_LL_D_HLIST_NODE_PTR(p);
181 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
182 spin_lock(&dentry->d_lock);
183 if (!list_empty(&dentry->d_subdirs)) {
184 struct dentry *child;
186 list_for_each_entry_safe(child, tmp_subdir,
189 if (child->d_inode == NULL)
190 d_lustre_invalidate(child, 1);
193 spin_unlock(&dentry->d_lock);
195 ll_unlock_dcache(dir);
198 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
199 void *data, int flag)
202 struct lustre_handle lockh;
206 case LDLM_CB_BLOCKING:
207 ldlm_lock2handle(lock, &lockh);
208 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
210 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
214 case LDLM_CB_CANCELING: {
215 struct inode *inode = ll_inode_from_resource_lock(lock);
216 struct ll_inode_info *lli;
217 __u64 bits = lock->l_policy_data.l_inodebits.bits;
219 ldlm_mode_t mode = lock->l_req_mode;
221 /* Inode is set to lock->l_resource->lr_lvb_inode
222 * for mdc - bug 24555 */
223 LASSERT(lock->l_ast_data == NULL);
225 /* Invalidate all dentries associated with this inode */
229 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
231 if (bits & MDS_INODELOCK_XATTR)
232 ll_xattr_cache_destroy(inode);
234 /* For OPEN locks we differentiate between lock modes
235 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
236 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
237 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
238 ll_have_md_lock(inode, &bits, LCK_MINMODE);
240 if (bits & MDS_INODELOCK_OPEN)
241 ll_have_md_lock(inode, &bits, mode);
243 fid = ll_inode2fid(inode);
244 if (!fid_res_name_eq(fid, &lock->l_resource->lr_name))
245 LDLM_ERROR(lock, "data mismatch with object "
246 DFID" (%p)", PFID(fid), inode);
248 if (bits & MDS_INODELOCK_OPEN) {
250 switch (lock->l_req_mode) {
261 CERROR("Unexpected lock mode for OPEN lock "
262 "%d, inode %ld\n", lock->l_req_mode,
265 ll_md_real_close(inode, flags);
268 lli = ll_i2info(inode);
269 if (bits & MDS_INODELOCK_LAYOUT) {
270 struct cl_object_conf conf = { { 0 } };
272 conf.coc_opc = OBJECT_CONF_INVALIDATE;
273 conf.coc_inode = inode;
274 rc = ll_layout_conf(inode, &conf);
276 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
279 if (bits & MDS_INODELOCK_UPDATE) {
280 spin_lock(&lli->lli_lock);
281 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
282 spin_unlock(&lli->lli_lock);
285 if (S_ISDIR(inode->i_mode) &&
286 (bits & MDS_INODELOCK_UPDATE)) {
287 CDEBUG(D_INODE, "invalidating inode %lu\n",
289 truncate_inode_pages(inode->i_mapping, 0);
290 ll_invalidate_negative_children(inode);
293 if (inode->i_sb->s_root &&
294 inode != inode->i_sb->s_root->d_inode &&
295 (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)))
296 ll_invalidate_aliases(inode);
307 __u32 ll_i2suppgid(struct inode *i)
309 if (in_group_p(i->i_gid))
310 return (__u32)i->i_gid;
315 /* Pack the required supplementary groups into the supplied groups array.
316 * If we don't need to use the groups from the target inode(s) then we
317 * instead pack one or more groups from the user's supplementary group
318 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
319 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
326 LASSERT(suppgids != NULL);
328 suppgids[0] = ll_i2suppgid(i1);
331 suppgids[1] = ll_i2suppgid(i2);
336 for (i = 0; i < current_ngroups; i++) {
337 if (suppgids[0] == -1) {
338 if (current_groups[i] != suppgids[1])
339 suppgids[0] = current_groups[i];
342 if (suppgids[1] == -1) {
343 if (current_groups[i] != suppgids[0])
344 suppgids[1] = current_groups[i];
353 * try to reuse three types of dentry:
354 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
355 * by concurrent .revalidate).
356 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
357 * be cleared by others calling d_lustre_revalidate).
358 * 3. DISCONNECTED alias.
360 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
362 struct dentry *alias, *discon_alias, *invalid_alias;
363 DECLARE_LL_D_HLIST_NODE_PTR(p);
365 if (ll_d_hlist_empty(&inode->i_dentry))
368 discon_alias = invalid_alias = NULL;
370 ll_lock_dcache(inode);
371 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
372 LASSERT(alias != dentry);
374 spin_lock(&alias->d_lock);
375 if (alias->d_flags & DCACHE_DISCONNECTED)
376 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
377 discon_alias = alias;
378 else if (alias->d_parent == dentry->d_parent &&
379 alias->d_name.hash == dentry->d_name.hash &&
380 alias->d_name.len == dentry->d_name.len &&
381 memcmp(alias->d_name.name, dentry->d_name.name,
382 dentry->d_name.len) == 0)
383 invalid_alias = alias;
384 spin_unlock(&alias->d_lock);
389 alias = invalid_alias ?: discon_alias ?: NULL;
391 spin_lock(&alias->d_lock);
393 spin_unlock(&alias->d_lock);
395 ll_unlock_dcache(inode);
401 * Similar to d_splice_alias(), but lustre treats invalid alias
402 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
404 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
410 new = ll_find_alias(inode, de);
420 "Reuse dentry %p inode %p refc %d flags %#x\n",
421 new, new->d_inode, d_refcount(new), new->d_flags);
429 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
430 de, de->d_inode, d_refcount(de), de->d_flags);
434 int ll_lookup_it_finish(struct ptlrpc_request *request,
435 struct lookup_intent *it, void *data)
437 struct it_cb_data *icbd = data;
438 struct dentry **de = icbd->icbd_childp;
439 struct inode *parent = icbd->icbd_parent;
440 struct inode *inode = NULL;
445 /* NB 1 request reference will be taken away by ll_intent_lock()
447 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
448 it->d.lustre.it_disposition);
449 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
450 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
454 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
456 /* We used to query real size from OSTs here, but actually
457 this is not needed. For stat() calls size would be updated
458 from subsequent do_revalidate()->ll_inode_revalidate_it() in
460 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
461 Everybody else who needs correct file size would call
462 ll_glimpse_size or some equivalent themselves anyway.
463 Also see bug 7198. */
466 /* Only hash *de if it is unhashed (new dentry).
467 * Atoimc_open may passin hashed dentries for open.
469 if (d_unhashed(*de)) {
470 *de = ll_splice_alias(inode, *de);
472 RETURN(PTR_ERR(*de));
475 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
476 /* we have lookup look - unhide dentry */
477 if (bits & MDS_INODELOCK_LOOKUP)
478 d_lustre_revalidate(*de);
479 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
480 /* If file created on server, don't depend on parent UPDATE
481 * lock to unhide it. It is left hidden and next lookup can
482 * find it in ll_splice_alias.
484 /* Check that parent has UPDATE lock. */
485 struct lookup_intent parent_it = {
487 .d.lustre.it_lock_handle = 0 };
489 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
490 &ll_i2info(parent)->lli_fid, NULL)) {
491 d_lustre_revalidate(*de);
492 ll_intent_release(&parent_it);
499 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
500 struct lookup_intent *it, int lookup_flags)
502 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
503 struct dentry *save = dentry, *retval;
504 struct ptlrpc_request *req = NULL;
505 struct md_op_data *op_data;
506 struct it_cb_data icbd;
511 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
512 RETURN(ERR_PTR(-ENAMETOOLONG));
514 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
515 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
516 parent->i_generation, parent, LL_IT2STR(it));
518 if (d_mountpoint(dentry))
519 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
521 ll_frob_intent(&it, &lookup_it);
523 if (it->it_op == IT_GETATTR) {
524 rc = ll_statahead_enter(parent, &dentry, 0);
527 GOTO(out, retval = NULL);
528 GOTO(out, retval = dentry);
532 icbd.icbd_childp = &dentry;
533 icbd.icbd_parent = parent;
535 if (it->it_op & IT_CREAT)
536 opc = LUSTRE_OPC_CREATE;
538 opc = LUSTRE_OPC_ANY;
540 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
541 dentry->d_name.len, lookup_flags, opc,
544 RETURN((void *)op_data);
546 /* enforce umask if acl disabled or MDS doesn't support umask */
547 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
548 it->it_create_mode &= ~current_umask();
550 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
551 lookup_flags, &req, ll_md_blocking_ast, 0);
552 ll_finish_md_op_data(op_data);
554 GOTO(out, retval = ERR_PTR(rc));
556 rc = ll_lookup_it_finish(req, it, &icbd);
558 ll_intent_release(it);
559 GOTO(out, retval = ERR_PTR(rc));
562 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
563 !S_ISREG(dentry->d_inode->i_mode) &&
564 !S_ISDIR(dentry->d_inode->i_mode)) {
565 ll_release_openhandle(dentry, it);
567 ll_lookup_finish_locks(it, dentry);
570 GOTO(out, retval = NULL);
572 GOTO(out, retval = dentry);
575 ptlrpc_req_finished(req);
576 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
577 ll_statahead_mark(parent, dentry);
581 #ifdef HAVE_IOP_ATOMIC_OPEN
582 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
585 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
588 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),flags=%u\n",
589 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
590 parent->i_generation, parent, flags);
592 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
593 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
596 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
600 de = ll_lookup_it(parent, dentry, itp, 0);
603 ll_intent_release(itp);
609 * For cached negative dentry and new dentry, handle lookup/create/open
612 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
613 struct file *file, unsigned open_flags,
614 umode_t mode, int *opened)
616 struct lookup_intent *it;
618 long long lookup_flags = LOOKUP_OPEN;
622 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),file %p,"
623 "open_flags %x,mode %x opened %d\n",
624 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
625 dir->i_generation, dir, file, open_flags, mode, *opened);
627 OBD_ALLOC(it, sizeof(*it));
632 if (open_flags & O_CREAT) {
633 it->it_op |= IT_CREAT;
634 lookup_flags |= LOOKUP_CREATE;
636 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
637 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
639 /* Dentry added to dcache tree in ll_lookup_it */
640 de = ll_lookup_it(dir, dentry, it, lookup_flags);
647 if (it_disposition(it, DISP_OPEN_CREATE)) {
648 /* Dentry instantiated in ll_create_it. */
649 rc = ll_create_it(dir, dentry, mode, it);
651 /* We dget in ll_splice_alias. */
657 *opened |= FILE_CREATED;
659 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
661 if (S_ISFIFO(dentry->d_inode->i_mode)) {
662 /* We cannot call open here as it would
665 if (it_disposition(it, DISP_ENQ_OPEN_REF))
667 (struct ptlrpc_request *)
668 it->d.lustre.it_data);
669 rc = finish_no_open(file, de);
671 file->private_data = it;
672 rc = finish_open(file, dentry, NULL, opened);
673 /* We dget in ll_splice_alias. finish_open takes
674 * care of dget for fd open.
680 rc = finish_no_open(file, de);
685 ll_intent_release(it);
686 OBD_FREE(it, sizeof(*it));
691 #else /* !HAVE_IOP_ATOMIC_OPEN */
692 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
695 struct lookup_intent *it;
697 OBD_ALLOC(it, sizeof(*it));
699 return ERR_PTR(-ENOMEM);
701 if (lookup_flags & LOOKUP_OPEN) {
703 if (lookup_flags & LOOKUP_CREATE)
704 it->it_op |= IT_CREAT;
705 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
706 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
708 it->it_op = IT_GETATTR;
714 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
715 struct nameidata *nd)
720 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
721 struct lookup_intent *it;
723 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
724 it = ll_d2d(dentry)->lld_it;
725 ll_d2d(dentry)->lld_it = NULL;
727 if ((nd->flags & LOOKUP_CREATE) &&
728 !(nd->flags & LOOKUP_OPEN))
731 it = ll_convert_intent(&nd->intent.open, nd->flags);
733 RETURN((struct dentry *)it);
736 de = ll_lookup_it(parent, dentry, it, nd->flags);
739 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
740 if (dentry->d_inode &&
741 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
742 if (S_ISFIFO(dentry->d_inode->i_mode)) {
743 // We cannot call open here as it would
746 (struct ptlrpc_request *)
747 it->d.lustre.it_data);
751 nd->intent.open.file->private_data = it;
752 filp = lookup_instantiate_filp(nd,
758 de = (struct dentry *)filp;
761 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
762 // XXX This can only reliably work on assumption
763 // that there are NO hashed negative dentries.
764 ll_d2d(dentry)->lld_it = it;
765 it = NULL; /* Will be freed in ll_create_nd */
766 /* We absolutely depend on ll_create_nd to be
767 * called to not leak this intent and possible
768 * data attached to it */
773 ll_intent_release(it);
774 OBD_FREE(it, sizeof(*it));
777 de = ll_lookup_it(parent, dentry, NULL, 0);
782 #endif /* HAVE_IOP_ATOMIC_OPEN */
784 /* We depend on "mode" being set with the proper file type/umask by now */
785 static struct inode *ll_create_node(struct inode *dir, const char *name,
786 int namelen, const void *data, int datalen,
787 int mode, __u64 extra,
788 struct lookup_intent *it)
790 struct inode *inode = NULL;
791 struct ptlrpc_request *request = NULL;
792 struct ll_sb_info *sbi = ll_i2sbi(dir);
796 LASSERT(it && it->d.lustre.it_disposition);
798 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
799 request = it->d.lustre.it_data;
800 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
801 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
803 GOTO(out, inode = ERR_PTR(rc));
805 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
807 /* We asked for a lock on the directory, but were granted a
808 * lock on the inode. Since we finally have an inode pointer,
809 * stuff it in the lock. */
810 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
811 inode, inode->i_ino, inode->i_generation);
812 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
815 ptlrpc_req_finished(request);
820 * By the time this is called, we already have created the directory cache
821 * entry for the new file, but it is so far negative - it has no inode.
823 * We defer creating the OBD object(s) until open, to keep the intent and
824 * non-intent code paths similar, and also because we do not have the MDS
825 * inode number before calling ll_create_node() (which is needed for LOV),
826 * so we would need to do yet another RPC to the MDS to store the LOV EA
827 * data on the MDS. If needed, we would pass the PACKED lmm as data and
828 * lmm_size in datalen (the MDS still has code which will handle that).
830 * If the create succeeds, we fill in the inode information
831 * with d_instantiate().
833 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
834 struct lookup_intent *it)
840 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
841 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
842 dir->i_generation, dir, LL_IT2STR(it));
844 rc = it_open_error(DISP_OPEN_CREATE, it);
848 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
849 NULL, 0, mode, 0, it);
851 RETURN(PTR_ERR(inode));
853 if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
854 ll_i2info(inode)->lli_volatile = true;
856 d_instantiate(dentry, inode);
860 static void ll_update_times(struct ptlrpc_request *request,
863 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
867 if (body->valid & OBD_MD_FLMTIME &&
868 body->mtime > LTIME_S(inode->i_mtime)) {
869 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
870 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
871 LTIME_S(inode->i_mtime) = body->mtime;
873 if (body->valid & OBD_MD_FLCTIME &&
874 body->ctime > LTIME_S(inode->i_ctime))
875 LTIME_S(inode->i_ctime) = body->ctime;
878 static int ll_new_node(struct inode *dir, struct qstr *name,
879 const char *tgt, int mode, int rdev,
880 struct dentry *dchild, __u32 opc)
882 struct ptlrpc_request *request = NULL;
883 struct md_op_data *op_data;
884 struct inode *inode = NULL;
885 struct ll_sb_info *sbi = ll_i2sbi(dir);
890 if (unlikely(tgt != NULL))
891 tgt_len = strlen(tgt) + 1;
893 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
894 name->len, 0, opc, NULL);
896 GOTO(err_exit, err = PTR_ERR(op_data));
898 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
899 current_fsuid(), current_fsgid(),
900 cfs_curproc_cap_pack(), rdev, &request);
901 ll_finish_md_op_data(op_data);
905 ll_update_times(request, dir);
908 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
912 d_instantiate(dchild, inode);
916 ptlrpc_req_finished(request);
921 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
922 unsigned rdev, struct dentry *dchild)
927 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
928 name->len, name->name, dir->i_ino, dir->i_generation, dir,
931 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
932 mode &= ~current_umask();
934 switch (mode & S_IFMT) {
936 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
942 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
953 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
958 #ifdef HAVE_IOP_ATOMIC_OPEN
960 * Plain create. Intent create is handled in atomic_open.
962 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
963 umode_t mode, bool want_excl)
967 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),"
968 "flags=%u, excl=%d\n",
969 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
970 dir->i_generation, dir, mode, want_excl);
972 rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
974 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
976 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
977 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
981 #else /* !HAVE_IOP_ATOMIC_OPEN */
982 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
983 ll_umode_t mode, struct nameidata *nd)
985 struct ll_dentry_data *lld = ll_d2d(dentry);
986 struct lookup_intent *it = NULL;
993 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
997 /* Was there an error? Propagate it! */
998 if (it->d.lustre.it_status) {
999 rc = it->d.lustre.it_status;
1003 rc = ll_create_it(dir, dentry, mode, it);
1004 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1007 nd->intent.open.file->private_data = it;
1008 filp = lookup_instantiate_filp(nd, dentry, NULL);
1014 ll_intent_release(it);
1015 OBD_FREE(it, sizeof(*it));
1018 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1022 #endif /* HAVE_IOP_ATOMIC_OPEN */
1024 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1025 const char *tgt, struct dentry *dchild)
1030 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
1031 name->len, name->name, dir->i_ino, dir->i_generation,
1034 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1035 0, dchild, LUSTRE_OPC_SYMLINK);
1038 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1043 static int ll_link_generic(struct inode *src, struct inode *dir,
1044 struct qstr *name, struct dentry *dchild)
1046 struct ll_sb_info *sbi = ll_i2sbi(dir);
1047 struct ptlrpc_request *request = NULL;
1048 struct md_op_data *op_data;
1053 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1054 src->i_ino, src->i_generation, src, dir->i_ino,
1055 dir->i_generation, dir, name->len, name->name);
1057 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1058 0, LUSTRE_OPC_ANY, NULL);
1059 if (IS_ERR(op_data))
1060 RETURN(PTR_ERR(op_data));
1062 err = md_link(sbi->ll_md_exp, op_data, &request);
1063 ll_finish_md_op_data(op_data);
1067 ll_update_times(request, dir);
1068 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1071 ptlrpc_req_finished(request);
1075 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1076 int mode, struct dentry *dchild)
1082 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1083 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1085 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1086 mode &= ~current_umask();
1087 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1088 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1091 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1096 /* Try to find the child dentry by its name.
1097 If found, put the result fid into @fid. */
1098 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1101 struct dentry *parent, *child;
1103 parent = ll_d_hlist_entry(dir->i_dentry, struct dentry, d_alias);
1104 child = d_lookup(parent, name);
1107 *fid = *ll_inode2fid(child->d_inode);
1112 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1113 struct dentry *dchild, struct qstr *name)
1115 struct ptlrpc_request *request = NULL;
1116 struct md_op_data *op_data;
1120 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1121 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1123 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1126 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1127 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1128 if (IS_ERR(op_data))
1129 RETURN(PTR_ERR(op_data));
1131 ll_get_child_fid(dir, name, &op_data->op_fid3);
1132 op_data->op_fid2 = op_data->op_fid3;
1133 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1134 ll_finish_md_op_data(op_data);
1136 ll_update_times(request, dir);
1137 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1140 ptlrpc_req_finished(request);
1147 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1149 struct ptlrpc_request *request = NULL;
1150 struct md_op_data *op_data;
1154 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1155 namelen, name, dir->i_ino, dir->i_generation, dir);
1157 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1158 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1159 if (IS_ERR(op_data))
1160 RETURN(PTR_ERR(op_data));
1161 op_data->op_cli_flags |= CLI_RM_ENTRY;
1162 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1163 ll_finish_md_op_data(op_data);
1165 ll_update_times(request, dir);
1166 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1169 ptlrpc_req_finished(request);
1173 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1175 struct mdt_body *body;
1176 struct lov_mds_md *eadata;
1177 struct lov_stripe_md *lsm = NULL;
1178 struct obd_trans_info oti = { 0 };
1180 struct obd_capa *oc = NULL;
1184 /* req is swabbed so this is safe */
1185 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1186 if (!(body->valid & OBD_MD_FLEASIZE))
1189 if (body->eadatasize == 0) {
1190 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1191 GOTO(out, rc = -EPROTO);
1194 /* The MDS sent back the EA because we unlinked the last reference
1195 * to this file. Use this EA to unlink the objects on the OST.
1196 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1197 * check it is complete and sensible. */
1198 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1200 LASSERT(eadata != NULL);
1202 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1204 CERROR("obd_unpackmd: %d\n", rc);
1207 LASSERT(rc >= sizeof(*lsm));
1211 GOTO(out_free_memmd, rc = -ENOMEM);
1213 oa->o_oi = lsm->lsm_oi;
1214 oa->o_mode = body->mode & S_IFMT;
1215 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1217 if (body->valid & OBD_MD_FLCOOKIE) {
1218 oa->o_valid |= OBD_MD_FLCOOKIE;
1219 oti.oti_logcookies =
1220 req_capsule_server_sized_get(&request->rq_pill,
1222 sizeof(struct llog_cookie) *
1223 lsm->lsm_stripe_count);
1224 if (oti.oti_logcookies == NULL) {
1225 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1226 body->valid &= ~OBD_MD_FLCOOKIE;
1230 if (body->valid & OBD_MD_FLOSSCAPA) {
1231 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1233 GOTO(out_free_memmd, rc);
1236 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1237 ll_i2mdexp(dir), oc);
1240 CERROR("obd destroy objid "DOSTID" error %d\n",
1241 POSTID(&lsm->lsm_oi), rc);
1243 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1249 /* ll_unlink_generic() doesn't update the inode with the new link count.
1250 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1251 * is any lock existing. They will recycle dentries and inodes based upon locks
1253 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1254 struct dentry *dchild, struct qstr *name)
1256 struct ptlrpc_request *request = NULL;
1257 struct md_op_data *op_data;
1260 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1261 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1264 * XXX: unlink bind mountpoint maybe call to here,
1265 * just check it as vfs_unlink does.
1267 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1270 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1271 name->len, 0, LUSTRE_OPC_ANY, NULL);
1272 if (IS_ERR(op_data))
1273 RETURN(PTR_ERR(op_data));
1275 ll_get_child_fid(dir, name, &op_data->op_fid3);
1276 op_data->op_fid2 = op_data->op_fid3;
1277 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1278 ll_finish_md_op_data(op_data);
1282 ll_update_times(request, dir);
1283 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1285 rc = ll_objects_destroy(request, dir);
1287 ptlrpc_req_finished(request);
1291 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1292 struct dentry *src_dchild, struct qstr *src_name,
1293 struct inode *tgt, struct dentry *tgt_dparent,
1294 struct dentry *tgt_dchild, struct qstr *tgt_name)
1296 struct ptlrpc_request *request = NULL;
1297 struct ll_sb_info *sbi = ll_i2sbi(src);
1298 struct md_op_data *op_data;
1301 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1302 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1303 src->i_ino, src->i_generation, src, tgt_name->len,
1304 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1306 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1307 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1310 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1311 LUSTRE_OPC_ANY, NULL);
1312 if (IS_ERR(op_data))
1313 RETURN(PTR_ERR(op_data));
1315 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1316 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1317 err = md_rename(sbi->ll_md_exp, op_data,
1318 src_name->name, src_name->len,
1319 tgt_name->name, tgt_name->len, &request);
1320 ll_finish_md_op_data(op_data);
1322 ll_update_times(request, src);
1323 ll_update_times(request, tgt);
1324 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1325 err = ll_objects_destroy(request, src);
1328 ptlrpc_req_finished(request);
1333 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1336 return ll_mknod_generic(dir, &dchild->d_name, mode,
1337 old_encode_dev(rdev), dchild);
1340 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1342 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1345 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1347 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1350 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1352 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1355 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1356 const char *oldname)
1358 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1361 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1362 struct dentry *new_dentry)
1364 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1368 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1369 struct inode *new_dir, struct dentry *new_dentry)
1372 err = ll_rename_generic(old_dir, NULL,
1373 old_dentry, &old_dentry->d_name,
1374 new_dir, NULL, new_dentry,
1375 &new_dentry->d_name);
1377 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1378 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1380 d_move(old_dentry, new_dentry);
1385 struct inode_operations ll_dir_inode_operations = {
1387 #ifdef HAVE_IOP_ATOMIC_OPEN
1388 .atomic_open = ll_atomic_open,
1390 .lookup = ll_lookup_nd,
1391 .create = ll_create_nd,
1392 /* We need all these non-raw things for NFSD, to not patch it. */
1393 .unlink = ll_unlink,
1396 .symlink = ll_symlink,
1398 .rename = ll_rename,
1399 .setattr = ll_setattr,
1400 .getattr = ll_getattr,
1401 .permission = ll_inode_permission,
1402 .setxattr = ll_setxattr,
1403 .getxattr = ll_getxattr,
1404 .listxattr = ll_listxattr,
1405 .removexattr = ll_removexattr,
1406 #ifdef HAVE_IOP_GET_ACL
1407 .get_acl = ll_get_acl,
1411 struct inode_operations ll_special_inode_operations = {
1412 .setattr = ll_setattr,
1413 .getattr = ll_getattr,
1414 .permission = ll_inode_permission,
1415 .setxattr = ll_setxattr,
1416 .getxattr = ll_getxattr,
1417 .listxattr = ll_listxattr,
1418 .removexattr = ll_removexattr,
1419 #ifdef HAVE_IOP_GET_ACL
1420 .get_acl = ll_get_acl,