1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
41 #include <linux/sched.h>
43 #include <linux/smp_lock.h>
44 #include <linux/quotaops.h>
45 #include <linux/highmem.h>
46 #include <linux/pagemap.h>
47 #include <linux/security.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include <obd_support.h>
52 #include <lustre_fid.h>
53 #include <lustre_lite.h>
54 #include <lustre_dlm.h>
55 #include <lustre_ver.h>
56 #include "llite_internal.h"
59 * Check if we have something mounted at the named dchild.
60 * In such a case there would always be dentry present.
62 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
67 if (unlikely(dchild)) {
68 mounted = d_mountpoint(dchild);
70 dchild = d_lookup(dparent, name);
72 mounted = d_mountpoint(dchild);
79 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
83 ldlm_lock_decref(lockh, mode);
89 /* called from iget5_locked->find_inode() under inode_lock spinlock */
90 static int ll_test_inode(struct inode *inode, void *opaque)
92 struct ll_inode_info *lli = ll_i2info(inode);
93 struct lustre_md *md = opaque;
95 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
96 CERROR("MDS body missing FID\n");
100 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
106 static int ll_set_inode(struct inode *inode, void *opaque)
108 struct ll_inode_info *lli = ll_i2info(inode);
109 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
111 if (unlikely(!(body->valid & OBD_MD_FLID))) {
112 CERROR("MDS body missing FID\n");
116 lli->lli_fid = body->fid1;
122 * Get an inode by inode number (already instantiated by the intent lookup).
123 * Returns inode or NULL
125 struct inode *ll_iget(struct super_block *sb, ino_t hash,
126 struct lustre_md *md)
132 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
135 if (inode->i_state & I_NEW) {
138 ll_read_inode2(inode, md);
139 if (S_ISREG(inode->i_mode) &&
140 ll_i2info(inode)->lli_clob == NULL)
141 rc = cl_inode_init(inode, md);
144 make_bad_inode(inode);
145 unlock_new_inode(inode);
149 unlock_new_inode(inode);
150 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
151 ll_update_inode(inode, md);
152 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
153 inode, PFID(&md->body->fid1));
158 static void ll_drop_negative_dentry(struct inode *dir)
160 struct dentry *dentry, *tmp_alias, *tmp_subdir;
162 cfs_spin_lock(&ll_lookup_lock);
163 spin_lock(&dcache_lock);
165 list_for_each_entry_safe(dentry, tmp_alias,
166 &dir->i_dentry,d_alias) {
167 if (!list_empty(&dentry->d_subdirs)) {
168 struct dentry *child;
169 list_for_each_entry_safe(child, tmp_subdir,
172 /* XXX Print some debug here? */
174 /* Negative dentry. If we were
175 dropping dcache lock, go
176 throught the list again */
177 if (ll_drop_dentry(child))
182 spin_unlock(&dcache_lock);
183 cfs_spin_unlock(&ll_lookup_lock);
187 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
188 void *data, int flag)
191 struct lustre_handle lockh;
195 case LDLM_CB_BLOCKING:
196 ldlm_lock2handle(lock, &lockh);
197 rc = ldlm_cli_cancel(&lockh);
199 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
203 case LDLM_CB_CANCELING: {
204 struct inode *inode = ll_inode_from_lock(lock);
205 struct ll_inode_info *lli;
206 __u64 bits = lock->l_policy_data.l_inodebits.bits;
208 ldlm_mode_t mode = lock->l_req_mode;
210 /* Invalidate all dentries associated with this inode */
214 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
215 /* For OPEN locks we differentiate between lock modes - CR, CW. PR - bug 22891 */
216 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE))
217 ll_have_md_lock(inode, &bits, LCK_MINMODE);
219 if (bits & MDS_INODELOCK_OPEN)
220 ll_have_md_lock(inode, &bits, mode);
222 fid = ll_inode2fid(inode);
223 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
224 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
225 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
226 LDLM_ERROR(lock, "data mismatch with object "
227 DFID" (%p)", PFID(fid), inode);
230 if (bits & MDS_INODELOCK_OPEN) {
232 switch (lock->l_req_mode) {
243 CERROR("Unexpected lock mode for OPEN lock "
244 "%d, inode %ld\n", lock->l_req_mode,
247 ll_md_real_close(inode, flags);
250 lli = ll_i2info(inode);
251 if (bits & MDS_INODELOCK_UPDATE)
252 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
254 if (S_ISDIR(inode->i_mode) &&
255 (bits & MDS_INODELOCK_UPDATE)) {
256 CDEBUG(D_INODE, "invalidating inode %lu\n",
258 truncate_inode_pages(inode->i_mapping, 0);
259 ll_drop_negative_dentry(inode);
262 if (inode->i_sb->s_root &&
263 inode != inode->i_sb->s_root->d_inode &&
264 (bits & MDS_INODELOCK_LOOKUP))
265 ll_unhash_aliases(inode);
276 __u32 ll_i2suppgid(struct inode *i)
278 if (cfs_curproc_is_in_groups(i->i_gid))
279 return (__u32)i->i_gid;
284 /* Pack the required supplementary groups into the supplied groups array.
285 * If we don't need to use the groups from the target inode(s) then we
286 * instead pack one or more groups from the user's supplementary group
287 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
288 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
295 LASSERT(suppgids != NULL);
297 suppgids[0] = ll_i2suppgid(i1);
300 suppgids[1] = ll_i2suppgid(i2);
305 for (i = 0; i < current_ngroups; i++) {
306 if (suppgids[0] == -1) {
307 if (current_groups[i] != suppgids[1])
308 suppgids[0] = current_groups[i];
311 if (suppgids[1] == -1) {
312 if (current_groups[i] != suppgids[0])
313 suppgids[1] = current_groups[i];
321 static void ll_d_add(struct dentry *de, struct inode *inode)
323 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
325 if (!list_empty(&de->d_alias)) {
326 spin_unlock(&dcache_lock);
327 CERROR("dentry %.*s %p alias next %p, prev %p\n",
328 de->d_name.len, de->d_name.name, de,
329 de->d_alias.next, de->d_alias.prev);
333 list_add(&de->d_alias, &inode->i_dentry);
335 /* d_instantiate() replacement code should initialize security
337 security_d_instantiate(de, inode);
340 if (!d_unhashed(de)) {
341 spin_unlock(&dcache_lock);
342 CERROR("dentry %.*s %p hash next %p\n",
343 de->d_name.len, de->d_name.name, de, de->d_hash.next);
346 d_rehash_cond(de, 0);
349 /* Search "inode"'s alias list for a dentry that has the same name and parent
350 * as de. If found, return it. If not found, return de.
351 * Lustre can't use d_add_unique because don't unhash aliases for directory
352 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
353 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
355 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
357 struct list_head *tmp;
358 struct dentry *dentry;
359 struct dentry *last_discon = NULL;
361 cfs_spin_lock(&ll_lookup_lock);
362 spin_lock(&dcache_lock);
363 list_for_each(tmp, &inode->i_dentry) {
364 dentry = list_entry(tmp, struct dentry, d_alias);
366 /* We are called here with 'de' already on the aliases list. */
367 if (unlikely(dentry == de)) {
372 if (dentry->d_flags & DCACHE_DISCONNECTED) {
373 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
374 last_discon = dentry;
378 if (dentry->d_parent != de->d_parent)
381 if (dentry->d_name.hash != de->d_name.hash)
384 if (dentry->d_name.len != de->d_name.len)
387 if (memcmp(dentry->d_name.name, de->d_name.name,
388 de->d_name.len) != 0)
392 ll_dops_init(dentry, 0, 1);
393 ll_dentry_rehash(dentry, 1);
394 spin_unlock(&dcache_lock);
395 cfs_spin_unlock(&ll_lookup_lock);
397 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
398 "refc %d\n", de->d_name.len, de->d_name.name, de,
399 de->d_parent, de->d_inode, atomic_read(&de->d_count));
404 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
405 "refc %d\n", last_discon, last_discon->d_inode,
406 atomic_read(&last_discon->d_count));
407 dget_locked(last_discon);
408 lock_dentry(last_discon);
409 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
410 unlock_dentry(last_discon);
411 spin_unlock(&dcache_lock);
412 cfs_spin_unlock(&ll_lookup_lock);
413 ll_dops_init(last_discon, 1, 1);
415 d_move(last_discon, de);
420 de->d_flags |= DCACHE_LUSTRE_INVALID;
424 spin_unlock(&dcache_lock);
425 cfs_spin_unlock(&ll_lookup_lock);
430 int ll_lookup_it_finish(struct ptlrpc_request *request,
431 struct lookup_intent *it, void *data)
433 struct it_cb_data *icbd = data;
434 struct dentry **de = icbd->icbd_childp;
435 struct inode *parent = icbd->icbd_parent;
436 struct inode *inode = NULL;
440 /* NB 1 request reference will be taken away by ll_intent_lock()
442 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
443 struct dentry *save = *de;
446 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
450 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
452 /* We used to query real size from OSTs here, but actually
453 this is not needed. For stat() calls size would be updated
454 from subsequent do_revalidate()->ll_inode_revalidate_it() in
456 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
457 Everybody else who needs correct file size would call
458 cl_glimpse_size or some equivalent themselves anyway.
459 Also see bug 7198. */
460 ll_dops_init(*de, 1, 1);
461 *de = ll_find_alias(inode, *de);
463 struct ll_dentry_data *lld = ll_d2d(*de);
465 /* just make sure the ll_dentry_data is ready */
466 if (unlikely(lld == NULL))
467 ll_dops_init(*de, 1, 1);
469 /* we have lookup look - unhide dentry */
470 ll_dentry_reset_flags(*de, bits);
474 ll_dops_init(*de, 1, 1);
475 /* Check that parent has UPDATE lock. If there is none, we
476 cannot afford to hash this dentry (done by ll_d_add) as it
477 might get picked up later when UPDATE lock will appear */
478 ibits = MDS_INODELOCK_UPDATE;
479 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE)) {
480 spin_lock(&dcache_lock);
482 spin_unlock(&dcache_lock);
484 /* negative lookup - and don't have update lock to
487 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
490 (*de)->d_inode = NULL;
491 /* We do not want to hash the dentry if don`t have a
492 * lock, but if this dentry is later used in d_move,
493 * we'd hit uninitialised list head d_hash, so we just
494 * do this to init d_hash field but leave dentry
495 * unhashed. (bug 10796). */
504 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
505 struct lookup_intent *it, int lookup_flags)
507 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
508 struct dentry *save = dentry, *retval;
509 struct ptlrpc_request *req = NULL;
510 struct md_op_data *op_data;
511 struct it_cb_data icbd;
516 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
517 RETURN(ERR_PTR(-ENAMETOOLONG));
519 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
520 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
521 parent->i_generation, parent, LL_IT2STR(it));
523 if (d_mountpoint(dentry))
524 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
526 ll_frob_intent(&it, &lookup_it);
528 /* As do_lookup is called before follow_mount, root dentry may be left
529 * not valid, revalidate it here. */
530 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
531 (it->it_op & (IT_OPEN | IT_CREAT))) {
532 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
533 MDS_INODELOCK_LOOKUP);
538 if (it->it_op == IT_GETATTR) {
539 rc = ll_statahead_enter(parent, &dentry, 0);
542 GOTO(out, retval = NULL);
543 GOTO(out, retval = dentry);
547 icbd.icbd_childp = &dentry;
548 icbd.icbd_parent = parent;
550 if (it->it_op & IT_CREAT ||
551 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
552 opc = LUSTRE_OPC_CREATE;
554 opc = LUSTRE_OPC_ANY;
556 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
557 dentry->d_name.len, lookup_flags, opc,
560 RETURN((void *)op_data);
562 it->it_create_mode &= ~cfs_curproc_umask();
564 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
565 lookup_flags, &req, ll_md_blocking_ast, 0);
566 ll_finish_md_op_data(op_data);
568 GOTO(out, retval = ERR_PTR(rc));
570 rc = ll_lookup_it_finish(req, it, &icbd);
572 ll_intent_release(it);
573 GOTO(out, retval = ERR_PTR(rc));
576 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
577 !S_ISREG(dentry->d_inode->i_mode) &&
578 !S_ISDIR(dentry->d_inode->i_mode)) {
579 ll_release_openhandle(dentry, it);
581 ll_lookup_finish_locks(it, dentry);
584 GOTO(out, retval = NULL);
586 GOTO(out, retval = dentry);
589 ptlrpc_req_finished(req);
590 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
591 ll_statahead_mark(parent, dentry);
595 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
598 struct lookup_intent *it;
600 OBD_ALLOC(it, sizeof(*it));
602 return ERR_PTR(-ENOMEM);
604 if (lookup_flags & LOOKUP_OPEN) {
606 if (lookup_flags & LOOKUP_CREATE)
607 it->it_op |= IT_CREAT;
608 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
609 it->it_flags = oit->flags;
611 it->it_op = IT_GETATTR;
614 #ifndef HAVE_FILE_IN_STRUCT_INTENT
615 /* Since there is no way to pass our intent to ll_file_open,
616 * just check the file is there. Actual open will be done
618 if (it->it_op & IT_OPEN)
619 it->it_op = IT_LOOKUP;
625 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
626 struct nameidata *nd)
631 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
632 struct lookup_intent *it;
634 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
635 /* Did we came here from failed revalidate just to propagate
637 if (nd->flags & LOOKUP_OPEN)
638 if (IS_ERR(nd->intent.open.file))
639 RETURN((struct dentry *)nd->intent.open.file);
641 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
642 it = ll_d2d(dentry)->lld_it;
643 ll_d2d(dentry)->lld_it = NULL;
645 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
646 /* We are sure this is new dentry, so we need to create
647 our private data and set the dentry ops */
648 ll_dops_init(dentry, 1, 1);
651 it = ll_convert_intent(&nd->intent.open, nd->flags);
653 RETURN((struct dentry *)it);
656 de = ll_lookup_it(parent, dentry, it, nd->flags);
659 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
660 if (dentry->d_inode &&
661 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
662 #ifdef HAVE_FILE_IN_STRUCT_INTENT
663 if (S_ISFIFO(dentry->d_inode->i_mode)) {
664 // We cannot call open here as it would
667 (struct ptlrpc_request *)
668 it->d.lustre.it_data);
670 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
671 /* 2.6.1[456] have a bug in open_namei() that forgets to check
672 * nd->intent.open.file for error, so we need to return it as lookup's result
675 nd->intent.open.file->private_data = it;
676 filp =lookup_instantiate_filp(nd,dentry,
681 de = (struct dentry *) filp;
684 nd->intent.open.file->private_data = it;
685 (void)lookup_instantiate_filp(nd,dentry,
690 #else /* HAVE_FILE_IN_STRUCT_INTENT */
691 /* Release open handle as we have no way to
692 * pass it to ll_file_open */
693 ll_release_openhandle(dentry, it);
694 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
695 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
696 // XXX This can only reliably work on assumption
697 // that there are NO hashed negative dentries.
698 ll_d2d(dentry)->lld_it = it;
699 it = NULL; /* Will be freed in ll_create_nd */
700 /* We absolutely depend on ll_create_nd to be
701 * called to not leak this intent and possible
702 * data attached to it */
707 ll_intent_release(it);
708 OBD_FREE(it, sizeof(*it));
711 de = ll_lookup_it(parent, dentry, NULL, 0);
717 /* We depend on "mode" being set with the proper file type/umask by now */
718 static struct inode *ll_create_node(struct inode *dir, const char *name,
719 int namelen, const void *data, int datalen,
720 int mode, __u64 extra,
721 struct lookup_intent *it)
723 struct inode *inode = NULL;
724 struct ptlrpc_request *request = NULL;
725 struct ll_sb_info *sbi = ll_i2sbi(dir);
729 LASSERT(it && it->d.lustre.it_disposition);
731 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
732 request = it->d.lustre.it_data;
733 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
734 rc = ll_prep_inode(&inode, request, dir->i_sb);
736 GOTO(out, inode = ERR_PTR(rc));
738 LASSERT(list_empty(&inode->i_dentry));
740 /* We asked for a lock on the directory, but were granted a
741 * lock on the inode. Since we finally have an inode pointer,
742 * stuff it in the lock. */
743 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
744 inode, inode->i_ino, inode->i_generation);
745 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
748 ptlrpc_req_finished(request);
753 * By the time this is called, we already have created the directory cache
754 * entry for the new file, but it is so far negative - it has no inode.
756 * We defer creating the OBD object(s) until open, to keep the intent and
757 * non-intent code paths similar, and also because we do not have the MDS
758 * inode number before calling ll_create_node() (which is needed for LOV),
759 * so we would need to do yet another RPC to the MDS to store the LOV EA
760 * data on the MDS. If needed, we would pass the PACKED lmm as data and
761 * lmm_size in datalen (the MDS still has code which will handle that).
763 * If the create succeeds, we fill in the inode information
764 * with d_instantiate().
766 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
767 struct lookup_intent *it)
773 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
774 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
775 dir->i_generation, dir, LL_IT2STR(it));
777 rc = it_open_error(DISP_OPEN_CREATE, it);
781 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
782 NULL, 0, mode, 0, it);
784 RETURN(PTR_ERR(inode));
786 d_instantiate(dentry, inode);
787 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
788 * but some callers, e.g. do_coredump, expect dentry to be hashed after
789 * successful create. Hash it here. */
790 spin_lock(&dcache_lock);
791 if (d_unhashed(dentry))
792 d_rehash_cond(dentry, 0);
793 spin_unlock(&dcache_lock);
797 static void ll_update_times(struct ptlrpc_request *request,
800 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
804 if (body->valid & OBD_MD_FLMTIME &&
805 body->mtime > LTIME_S(inode->i_mtime)) {
806 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
807 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
808 LTIME_S(inode->i_mtime) = body->mtime;
810 if (body->valid & OBD_MD_FLCTIME &&
811 body->ctime > LTIME_S(inode->i_ctime))
812 LTIME_S(inode->i_ctime) = body->ctime;
815 static int ll_new_node(struct inode *dir, struct qstr *name,
816 const char *tgt, int mode, int rdev,
817 struct dentry *dchild, __u32 opc)
819 struct ptlrpc_request *request = NULL;
820 struct md_op_data *op_data;
821 struct inode *inode = NULL;
822 struct ll_sb_info *sbi = ll_i2sbi(dir);
827 if (unlikely(tgt != NULL))
828 tgt_len = strlen(tgt) + 1;
830 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
831 name->len, 0, opc, NULL);
833 GOTO(err_exit, err = PTR_ERR(op_data));
835 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
836 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
837 cfs_curproc_cap_pack(), rdev, &request);
838 ll_finish_md_op_data(op_data);
842 ll_update_times(request, dir);
845 err = ll_prep_inode(&inode, request, dchild->d_sb);
850 d_instantiate(dchild, inode);
854 ptlrpc_req_finished(request);
859 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
860 unsigned rdev, struct dentry *dchild)
865 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
866 name->len, name->name, dir->i_ino, dir->i_generation, dir,
869 mode &= ~cfs_curproc_umask();
871 switch (mode & S_IFMT) {
873 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
879 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
891 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
892 int mode, struct nameidata *nd)
894 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
898 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
900 ll_d2d(dentry)->lld_it = NULL;
902 /* Was there an error? Propagate it! */
903 if (it->d.lustre.it_status) {
904 rc = it->d.lustre.it_status;
908 rc = ll_create_it(dir, dentry, mode, it);
909 #ifdef HAVE_FILE_IN_STRUCT_INTENT
910 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
911 nd->intent.open.file->private_data = it;
912 lookup_instantiate_filp(nd, dentry, NULL);
915 ll_release_openhandle(dentry,it);
919 ll_intent_release(it);
920 OBD_FREE(it, sizeof(*it));
925 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
926 const char *tgt, struct dentry *dchild)
931 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
932 name->len, name->name, dir->i_ino, dir->i_generation,
935 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
936 0, dchild, LUSTRE_OPC_SYMLINK);
940 static int ll_link_generic(struct inode *src, struct inode *dir,
941 struct qstr *name, struct dentry *dchild)
943 struct ll_sb_info *sbi = ll_i2sbi(dir);
944 struct ptlrpc_request *request = NULL;
945 struct md_op_data *op_data;
950 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
951 src->i_ino, src->i_generation, src, dir->i_ino,
952 dir->i_generation, dir, name->len, name->name);
954 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
955 0, LUSTRE_OPC_ANY, NULL);
957 RETURN(PTR_ERR(op_data));
959 err = md_link(sbi->ll_md_exp, op_data, &request);
960 ll_finish_md_op_data(op_data);
966 ll_update_times(request, dir);
969 ptlrpc_req_finished(request);
973 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
974 int mode, struct dentry *dchild)
980 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
981 name->len, name->name, dir->i_ino, dir->i_generation, dir);
983 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
984 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
989 /* Try to find the child dentry by its name.
990 If found, put the result fid into @fid. */
991 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
994 struct dentry *parent, *child;
996 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
997 child = d_lookup(parent, name);
1000 *fid = *ll_inode2fid(child->d_inode);
1005 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1006 struct dentry *dchild, struct qstr *name)
1008 struct ptlrpc_request *request = NULL;
1009 struct md_op_data *op_data;
1013 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1014 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1016 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1019 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1020 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1021 if (IS_ERR(op_data))
1022 RETURN(PTR_ERR(op_data));
1024 ll_get_child_fid(dir, name, &op_data->op_fid3);
1025 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1026 ll_finish_md_op_data(op_data);
1028 ll_update_times(request, dir);
1029 ptlrpc_req_finished(request);
1033 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1035 struct mdt_body *body;
1036 struct lov_mds_md *eadata;
1037 struct lov_stripe_md *lsm = NULL;
1038 struct obd_trans_info oti = { 0 };
1040 struct obd_capa *oc = NULL;
1044 /* req is swabbed so this is safe */
1045 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1046 if (!(body->valid & OBD_MD_FLEASIZE))
1049 if (body->eadatasize == 0) {
1050 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1051 GOTO(out, rc = -EPROTO);
1054 /* The MDS sent back the EA because we unlinked the last reference
1055 * to this file. Use this EA to unlink the objects on the OST.
1056 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1057 * check it is complete and sensible. */
1058 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1060 LASSERT(eadata != NULL);
1062 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1064 CERROR("obd_unpackmd: %d\n", rc);
1067 LASSERT(rc >= sizeof(*lsm));
1071 GOTO(out_free_memmd, rc = -ENOMEM);
1073 oa->o_id = lsm->lsm_object_id;
1074 oa->o_seq = lsm->lsm_object_seq;
1075 oa->o_mode = body->mode & S_IFMT;
1076 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1078 if (body->valid & OBD_MD_FLCOOKIE) {
1079 oa->o_valid |= OBD_MD_FLCOOKIE;
1080 oti.oti_logcookies =
1081 req_capsule_server_sized_get(&request->rq_pill,
1083 sizeof(struct llog_cookie) *
1084 lsm->lsm_stripe_count);
1085 if (oti.oti_logcookies == NULL) {
1086 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1087 body->valid &= ~OBD_MD_FLCOOKIE;
1091 if (body->valid & OBD_MD_FLOSSCAPA) {
1092 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1094 GOTO(out_free_memmd, rc);
1097 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1101 CERROR("obd destroy objid "LPX64" error %d\n",
1102 lsm->lsm_object_id, rc);
1104 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1109 /* ll_unlink_generic() doesn't update the inode with the new link count.
1110 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1111 * is any lock existing. They will recycle dentries and inodes based upon locks
1113 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1114 struct dentry *dchild, struct qstr *name)
1116 struct ptlrpc_request *request = NULL;
1117 struct md_op_data *op_data;
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);
1124 * XXX: unlink bind mountpoint maybe call to here,
1125 * just check it as vfs_unlink does.
1127 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1130 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1131 name->len, 0, LUSTRE_OPC_ANY, NULL);
1132 if (IS_ERR(op_data))
1133 RETURN(PTR_ERR(op_data));
1135 ll_get_child_fid(dir, name, &op_data->op_fid3);
1136 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1137 ll_finish_md_op_data(op_data);
1141 ll_update_times(request, dir);
1143 rc = ll_objects_destroy(request, dir);
1145 ptlrpc_req_finished(request);
1149 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1150 struct dentry *src_dchild, struct qstr *src_name,
1151 struct inode *tgt, struct dentry *tgt_dparent,
1152 struct dentry *tgt_dchild, struct qstr *tgt_name)
1154 struct ptlrpc_request *request = NULL;
1155 struct ll_sb_info *sbi = ll_i2sbi(src);
1156 struct md_op_data *op_data;
1159 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1160 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1161 src->i_ino, src->i_generation, src, tgt_name->len,
1162 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1164 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1165 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1168 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1169 LUSTRE_OPC_ANY, NULL);
1170 if (IS_ERR(op_data))
1171 RETURN(PTR_ERR(op_data));
1173 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1174 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1175 err = md_rename(sbi->ll_md_exp, op_data,
1176 src_name->name, src_name->len,
1177 tgt_name->name, tgt_name->len, &request);
1178 ll_finish_md_op_data(op_data);
1180 ll_update_times(request, src);
1181 ll_update_times(request, tgt);
1182 err = ll_objects_destroy(request, src);
1185 ptlrpc_req_finished(request);
1190 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1193 return ll_mknod_generic(dir, &dchild->d_name, mode,
1194 old_encode_dev(rdev), dchild);
1197 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1199 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1201 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1203 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1205 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1207 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1209 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1210 const char *oldname)
1212 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1214 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1215 struct dentry *new_dentry)
1217 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1220 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1221 struct inode *new_dir, struct dentry *new_dentry)
1224 err = ll_rename_generic(old_dir, NULL,
1225 old_dentry, &old_dentry->d_name,
1226 new_dir, NULL, new_dentry,
1227 &new_dentry->d_name);
1229 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1230 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1232 d_move(old_dentry, new_dentry);
1237 struct inode_operations ll_dir_inode_operations = {
1239 .lookup = ll_lookup_nd,
1240 .create = ll_create_nd,
1241 /* We need all these non-raw things for NFSD, to not patch it. */
1242 .unlink = ll_unlink,
1245 .symlink = ll_symlink,
1247 .rename = ll_rename,
1248 .setattr = ll_setattr,
1249 .getattr = ll_getattr,
1250 .permission = ll_inode_permission,
1251 .setxattr = ll_setxattr,
1252 .getxattr = ll_getxattr,
1253 .listxattr = ll_listxattr,
1254 .removexattr = ll_removexattr,
1257 struct inode_operations ll_special_inode_operations = {
1258 .setattr = ll_setattr,
1259 .getattr = ll_getattr,
1260 .permission = ll_inode_permission,
1261 .setxattr = ll_setxattr,
1262 .getxattr = ll_getxattr,
1263 .listxattr = ll_listxattr,
1264 .removexattr = ll_removexattr,