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>
48 #define DEBUG_SUBSYSTEM S_LLITE
50 #include <obd_support.h>
51 #include <lustre_fid.h>
52 #include <lustre_lite.h>
53 #include <lustre_dlm.h>
54 #include <lustre_ver.h>
55 #include "llite_internal.h"
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;
121 * Get an inode by inode number (already instantiated by the intent lookup).
122 * Returns inode or NULL
124 struct inode *ll_iget(struct super_block *sb, ino_t hash,
125 struct lustre_md *md)
131 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
134 if (inode->i_state & I_NEW) {
137 ll_read_inode2(inode, md);
138 rc = cl_inode_init(inode, md);
141 make_bad_inode(inode);
142 unlock_new_inode(inode);
146 unlock_new_inode(inode);
147 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
148 ll_update_inode(inode, md);
149 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
150 inode, PFID(&md->body->fid1));
155 static void ll_drop_negative_dentry(struct inode *dir)
157 struct dentry *dentry, *tmp_alias, *tmp_subdir;
159 cfs_spin_lock(&ll_lookup_lock);
160 spin_lock(&dcache_lock);
162 list_for_each_entry_safe(dentry, tmp_alias,
163 &dir->i_dentry,d_alias) {
164 if (!list_empty(&dentry->d_subdirs)) {
165 struct dentry *child;
166 list_for_each_entry_safe(child, tmp_subdir,
169 /* XXX Print some debug here? */
171 /* Negative dentry. If we were
172 dropping dcache lock, go
173 throught the list again */
174 if (ll_drop_dentry(child))
179 spin_unlock(&dcache_lock);
180 cfs_spin_unlock(&ll_lookup_lock);
184 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
185 void *data, int flag)
188 struct lustre_handle lockh;
192 case LDLM_CB_BLOCKING:
193 ldlm_lock2handle(lock, &lockh);
194 rc = ldlm_cli_cancel(&lockh);
196 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
200 case LDLM_CB_CANCELING: {
201 struct inode *inode = ll_inode_from_lock(lock);
202 struct ll_inode_info *lli;
203 __u64 bits = lock->l_policy_data.l_inodebits.bits;
205 ldlm_mode_t mode = lock->l_req_mode;
207 /* Invalidate all dentries associated with this inode */
211 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
212 /* For OPEN locks we differentiate between lock modes - CR, CW. PR - bug 22891 */
213 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE))
214 ll_have_md_lock(inode, &bits, LCK_MINMODE);
216 if (bits & MDS_INODELOCK_OPEN)
217 ll_have_md_lock(inode, &bits, mode);
219 fid = ll_inode2fid(inode);
220 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
221 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
222 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
223 LDLM_ERROR(lock, "data mismatch with object "
224 DFID" (%p)", PFID(fid), inode);
227 if (bits & MDS_INODELOCK_OPEN) {
229 switch (lock->l_req_mode) {
240 CERROR("Unexpected lock mode for OPEN lock "
241 "%d, inode %ld\n", lock->l_req_mode,
244 ll_md_real_close(inode, flags);
247 lli = ll_i2info(inode);
248 if (bits & MDS_INODELOCK_UPDATE)
249 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
251 if (S_ISDIR(inode->i_mode) &&
252 (bits & MDS_INODELOCK_UPDATE)) {
253 CDEBUG(D_INODE, "invalidating inode %lu\n",
255 truncate_inode_pages(inode->i_mapping, 0);
256 ll_drop_negative_dentry(inode);
259 if (inode->i_sb->s_root &&
260 inode != inode->i_sb->s_root->d_inode &&
261 (bits & MDS_INODELOCK_LOOKUP))
262 ll_unhash_aliases(inode);
273 __u32 ll_i2suppgid(struct inode *i)
275 if (cfs_curproc_is_in_groups(i->i_gid))
276 return (__u32)i->i_gid;
281 /* Pack the required supplementary groups into the supplied groups array.
282 * If we don't need to use the groups from the target inode(s) then we
283 * instead pack one or more groups from the user's supplementary group
284 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
285 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
292 LASSERT(suppgids != NULL);
294 suppgids[0] = ll_i2suppgid(i1);
297 suppgids[1] = ll_i2suppgid(i2);
302 for (i = 0; i < current_ngroups; i++) {
303 if (suppgids[0] == -1) {
304 if (current_groups[i] != suppgids[1])
305 suppgids[0] = current_groups[i];
308 if (suppgids[1] == -1) {
309 if (current_groups[i] != suppgids[0])
310 suppgids[1] = current_groups[i];
318 static void ll_d_add(struct dentry *de, struct inode *inode)
320 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
322 if (!list_empty(&de->d_alias)) {
323 spin_unlock(&dcache_lock);
324 CERROR("dentry %.*s %p alias next %p, prev %p\n",
325 de->d_name.len, de->d_name.name, de,
326 de->d_alias.next, de->d_alias.prev);
330 list_add(&de->d_alias, &inode->i_dentry);
334 if (!d_unhashed(de)) {
335 spin_unlock(&dcache_lock);
336 CERROR("dentry %.*s %p hash next %p\n",
337 de->d_name.len, de->d_name.name, de, de->d_hash.next);
340 d_rehash_cond(de, 0);
343 /* Search "inode"'s alias list for a dentry that has the same name and parent
344 * as de. If found, return it. If not found, return de.
345 * Lustre can't use d_add_unique because don't unhash aliases for directory
346 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
347 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
349 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
351 struct list_head *tmp;
352 struct dentry *dentry;
353 struct dentry *last_discon = NULL;
355 cfs_spin_lock(&ll_lookup_lock);
356 spin_lock(&dcache_lock);
357 list_for_each(tmp, &inode->i_dentry) {
358 dentry = list_entry(tmp, struct dentry, d_alias);
360 /* We are called here with 'de' already on the aliases list. */
361 if (unlikely(dentry == de)) {
366 if (dentry->d_flags & DCACHE_DISCONNECTED) {
367 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
368 last_discon = dentry;
372 if (dentry->d_parent != de->d_parent)
375 if (dentry->d_name.hash != de->d_name.hash)
378 if (dentry->d_name.len != de->d_name.len)
381 if (memcmp(dentry->d_name.name, de->d_name.name,
382 de->d_name.len) != 0)
388 unlock_dentry(dentry);
389 ll_dops_init(dentry, 0, 1);
390 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
391 spin_unlock(&dcache_lock);
392 cfs_spin_unlock(&ll_lookup_lock);
394 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
395 "refc %d\n", de->d_name.len, de->d_name.name, de,
396 de->d_parent, de->d_inode, atomic_read(&de->d_count));
401 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
402 "refc %d\n", last_discon, last_discon->d_inode,
403 atomic_read(&last_discon->d_count));
404 dget_locked(last_discon);
405 lock_dentry(last_discon);
406 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
407 unlock_dentry(last_discon);
408 spin_unlock(&dcache_lock);
409 cfs_spin_unlock(&ll_lookup_lock);
410 ll_dops_init(last_discon, 1, 1);
412 d_move(last_discon, de);
417 de->d_flags |= DCACHE_LUSTRE_INVALID;
421 spin_unlock(&dcache_lock);
422 cfs_spin_unlock(&ll_lookup_lock);
427 int ll_lookup_it_finish(struct ptlrpc_request *request,
428 struct lookup_intent *it, void *data)
430 struct it_cb_data *icbd = data;
431 struct dentry **de = icbd->icbd_childp;
432 struct inode *parent = icbd->icbd_parent;
433 struct ll_sb_info *sbi = ll_i2sbi(parent);
434 struct inode *inode = NULL;
438 /* NB 1 request reference will be taken away by ll_intent_lock()
440 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
441 struct dentry *save = *de;
444 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
448 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
449 inode, inode->i_ino, inode->i_generation);
450 md_set_lock_data(sbi->ll_md_exp,
451 &it->d.lustre.it_lock_handle, inode, &bits);
453 /* We used to query real size from OSTs here, but actually
454 this is not needed. For stat() calls size would be updated
455 from subsequent do_revalidate()->ll_inode_revalidate_it() in
457 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
458 Everybody else who needs correct file size would call
459 cl_glimpse_size or some equivalent themselves anyway.
460 Also see bug 7198. */
461 ll_dops_init(*de, 1, 1);
462 *de = ll_find_alias(inode, *de);
464 struct ll_dentry_data *lld = ll_d2d(*de);
466 /* just make sure the ll_dentry_data is ready */
467 if (unlikely(lld == NULL))
468 ll_dops_init(*de, 1, 1);
470 /* we have lookup look - unhide dentry */
471 if (bits & MDS_INODELOCK_LOOKUP) {
473 (*de)->d_flags &= ~DCACHE_LUSTRE_INVALID;
479 ll_dops_init(*de, 1, 1);
480 /* Check that parent has UPDATE lock. If there is none, we
481 cannot afford to hash this dentry (done by ll_d_add) as it
482 might get picked up later when UPDATE lock will appear */
483 ibits = MDS_INODELOCK_UPDATE;
484 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE)) {
485 spin_lock(&dcache_lock);
487 spin_unlock(&dcache_lock);
489 /* negative lookup - and don't have update lock to
492 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
495 (*de)->d_inode = NULL;
496 /* We do not want to hash the dentry if don`t have a
497 * lock, but if this dentry is later used in d_move,
498 * we'd hit uninitialised list head d_hash, so we just
499 * do this to init d_hash field but leave dentry
500 * unhashed. (bug 10796). */
509 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
510 struct lookup_intent *it, int lookup_flags)
512 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
513 struct dentry *save = dentry, *retval;
514 struct ptlrpc_request *req = NULL;
515 struct md_op_data *op_data;
516 struct it_cb_data icbd;
521 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
522 RETURN(ERR_PTR(-ENAMETOOLONG));
524 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
525 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
526 parent->i_generation, parent, LL_IT2STR(it));
528 if (d_mountpoint(dentry))
529 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
531 ll_frob_intent(&it, &lookup_it);
533 /* As do_lookup is called before follow_mount, root dentry may be left
534 * not valid, revalidate it here. */
535 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
536 (it->it_op & (IT_OPEN | IT_CREAT))) {
537 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
538 MDS_INODELOCK_LOOKUP);
543 if (it->it_op == IT_GETATTR) {
544 first = ll_statahead_enter(parent, &dentry, 1);
546 ll_statahead_exit(parent, dentry, first);
548 RETURN(retval = dentry);
552 icbd.icbd_childp = &dentry;
553 icbd.icbd_parent = parent;
555 if (it->it_op & IT_CREAT ||
556 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
557 opc = LUSTRE_OPC_CREATE;
559 opc = LUSTRE_OPC_ANY;
561 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
562 dentry->d_name.len, lookup_flags, opc,
565 RETURN((void *)op_data);
567 it->it_create_mode &= ~cfs_curproc_umask();
569 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
570 lookup_flags, &req, ll_md_blocking_ast, 0);
571 ll_finish_md_op_data(op_data);
573 GOTO(out, retval = ERR_PTR(rc));
575 rc = ll_lookup_it_finish(req, it, &icbd);
577 ll_intent_release(it);
578 GOTO(out, retval = ERR_PTR(rc));
581 if (first == -EEXIST)
582 ll_statahead_mark(parent, dentry);
584 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
585 !S_ISREG(dentry->d_inode->i_mode) &&
586 !S_ISDIR(dentry->d_inode->i_mode)) {
587 ll_release_openhandle(dentry, it);
589 ll_lookup_finish_locks(it, dentry);
592 GOTO(out, retval = NULL);
594 GOTO(out, retval = dentry);
597 ptlrpc_req_finished(req);
601 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
604 struct lookup_intent *it;
606 OBD_ALLOC(it, sizeof(*it));
608 return ERR_PTR(-ENOMEM);
610 if (lookup_flags & LOOKUP_OPEN) {
612 if (lookup_flags & LOOKUP_CREATE)
613 it->it_op |= IT_CREAT;
614 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
615 it->it_flags = oit->flags;
617 it->it_op = IT_GETATTR;
620 #ifndef HAVE_FILE_IN_STRUCT_INTENT
621 /* Since there is no way to pass our intent to ll_file_open,
622 * just check the file is there. Actual open will be done
624 if (it->it_op & IT_OPEN)
625 it->it_op = IT_LOOKUP;
631 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
632 struct nameidata *nd)
637 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
638 struct lookup_intent *it;
640 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
641 /* Did we came here from failed revalidate just to propagate
643 if (nd->flags & LOOKUP_OPEN)
644 if (IS_ERR(nd->intent.open.file))
645 RETURN((struct dentry *)nd->intent.open.file);
647 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
648 it = ll_d2d(dentry)->lld_it;
649 ll_d2d(dentry)->lld_it = NULL;
651 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
652 /* We are sure this is new dentry, so we need to create
653 our private data and set the dentry ops */
654 ll_dops_init(dentry, 1, 1);
657 it = ll_convert_intent(&nd->intent.open, nd->flags);
659 RETURN((struct dentry *)it);
662 de = ll_lookup_it(parent, dentry, it, nd->flags);
665 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
666 if (dentry->d_inode &&
667 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
668 #ifdef HAVE_FILE_IN_STRUCT_INTENT
669 if (S_ISFIFO(dentry->d_inode->i_mode)) {
670 // We cannot call open here as it would
673 (struct ptlrpc_request *)
674 it->d.lustre.it_data);
676 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
677 /* 2.6.1[456] have a bug in open_namei() that forgets to check
678 * nd->intent.open.file for error, so we need to return it as lookup's result
681 nd->intent.open.file->private_data = it;
682 filp =lookup_instantiate_filp(nd,dentry,
687 de = (struct dentry *) filp;
690 nd->intent.open.file->private_data = it;
691 (void)lookup_instantiate_filp(nd,dentry,
696 #else /* HAVE_FILE_IN_STRUCT_INTENT */
697 /* Release open handle as we have no way to
698 * pass it to ll_file_open */
699 ll_release_openhandle(dentry, it);
700 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
701 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
702 // XXX This can only reliably work on assumption
703 // that there are NO hashed negative dentries.
704 ll_d2d(dentry)->lld_it = it;
705 it = NULL; /* Will be freed in ll_create_nd */
706 /* We absolutely depend on ll_create_nd to be
707 * called to not leak this intent and possible
708 * data attached to it */
713 ll_intent_release(it);
714 OBD_FREE(it, sizeof(*it));
717 de = ll_lookup_it(parent, dentry, NULL, 0);
723 /* We depend on "mode" being set with the proper file type/umask by now */
724 static struct inode *ll_create_node(struct inode *dir, const char *name,
725 int namelen, const void *data, int datalen,
726 int mode, __u64 extra,
727 struct lookup_intent *it)
729 struct inode *inode = NULL;
730 struct ptlrpc_request *request = NULL;
731 struct ll_sb_info *sbi = ll_i2sbi(dir);
735 LASSERT(it && it->d.lustre.it_disposition);
737 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
738 request = it->d.lustre.it_data;
739 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
740 rc = ll_prep_inode(&inode, request, dir->i_sb);
742 GOTO(out, inode = ERR_PTR(rc));
744 LASSERT(list_empty(&inode->i_dentry));
746 /* We asked for a lock on the directory, but were granted a
747 * lock on the inode. Since we finally have an inode pointer,
748 * stuff it in the lock. */
749 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
750 inode, inode->i_ino, inode->i_generation);
751 md_set_lock_data(sbi->ll_md_exp,
752 &it->d.lustre.it_lock_handle, inode, NULL);
755 ptlrpc_req_finished(request);
760 * By the time this is called, we already have created the directory cache
761 * entry for the new file, but it is so far negative - it has no inode.
763 * We defer creating the OBD object(s) until open, to keep the intent and
764 * non-intent code paths similar, and also because we do not have the MDS
765 * inode number before calling ll_create_node() (which is needed for LOV),
766 * so we would need to do yet another RPC to the MDS to store the LOV EA
767 * data on the MDS. If needed, we would pass the PACKED lmm as data and
768 * lmm_size in datalen (the MDS still has code which will handle that).
770 * If the create succeeds, we fill in the inode information
771 * with d_instantiate().
773 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
774 struct lookup_intent *it)
780 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
781 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
782 dir->i_generation, dir, LL_IT2STR(it));
784 rc = it_open_error(DISP_OPEN_CREATE, it);
788 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
789 NULL, 0, mode, 0, it);
791 RETURN(PTR_ERR(inode));
793 d_instantiate(dentry, inode);
794 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
795 * but some callers, e.g. do_coredump, expect dentry to be hashed after
796 * successful create. Hash it here. */
797 spin_lock(&dcache_lock);
798 if (d_unhashed(dentry))
799 d_rehash_cond(dentry, 0);
800 spin_unlock(&dcache_lock);
804 static void ll_update_times(struct ptlrpc_request *request,
807 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
811 if (body->valid & OBD_MD_FLMTIME &&
812 body->mtime > LTIME_S(inode->i_mtime)) {
813 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
814 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
815 LTIME_S(inode->i_mtime) = body->mtime;
817 if (body->valid & OBD_MD_FLCTIME &&
818 body->ctime > LTIME_S(inode->i_ctime))
819 LTIME_S(inode->i_ctime) = body->ctime;
822 static int ll_new_node(struct inode *dir, struct qstr *name,
823 const char *tgt, int mode, int rdev,
824 struct dentry *dchild, __u32 opc)
826 struct ptlrpc_request *request = NULL;
827 struct md_op_data *op_data;
828 struct inode *inode = NULL;
829 struct ll_sb_info *sbi = ll_i2sbi(dir);
834 if (unlikely(tgt != NULL))
835 tgt_len = strlen(tgt) + 1;
837 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
838 name->len, 0, opc, NULL);
840 GOTO(err_exit, err = PTR_ERR(op_data));
842 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
843 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
844 cfs_curproc_cap_pack(), rdev, &request);
845 ll_finish_md_op_data(op_data);
849 ll_update_times(request, dir);
852 err = ll_prep_inode(&inode, request, dchild->d_sb);
857 d_instantiate(dchild, inode);
861 ptlrpc_req_finished(request);
866 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
867 unsigned rdev, struct dentry *dchild)
872 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
873 name->len, name->name, dir->i_ino, dir->i_generation, dir,
876 mode &= ~cfs_curproc_umask();
878 switch (mode & S_IFMT) {
880 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
886 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
898 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
899 int mode, struct nameidata *nd)
901 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
905 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
907 ll_d2d(dentry)->lld_it = NULL;
909 /* Was there an error? Propagate it! */
910 if (it->d.lustre.it_status) {
911 rc = it->d.lustre.it_status;
915 rc = ll_create_it(dir, dentry, mode, it);
916 #ifdef HAVE_FILE_IN_STRUCT_INTENT
917 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
918 nd->intent.open.file->private_data = it;
919 lookup_instantiate_filp(nd, dentry, NULL);
922 ll_release_openhandle(dentry,it);
926 ll_intent_release(it);
927 OBD_FREE(it, sizeof(*it));
932 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
933 const char *tgt, struct dentry *dchild)
938 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
939 name->len, name->name, dir->i_ino, dir->i_generation,
942 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
943 0, dchild, LUSTRE_OPC_SYMLINK);
947 static int ll_link_generic(struct inode *src, struct inode *dir,
948 struct qstr *name, struct dentry *dchild)
950 struct ll_sb_info *sbi = ll_i2sbi(dir);
951 struct ptlrpc_request *request = NULL;
952 struct md_op_data *op_data;
957 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
958 src->i_ino, src->i_generation, src, dir->i_ino,
959 dir->i_generation, dir, name->len, name->name);
961 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
962 0, LUSTRE_OPC_ANY, NULL);
964 RETURN(PTR_ERR(op_data));
966 err = md_link(sbi->ll_md_exp, op_data, &request);
967 ll_finish_md_op_data(op_data);
973 ll_update_times(request, dir);
976 ptlrpc_req_finished(request);
980 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
981 int mode, struct dentry *dchild)
987 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
988 name->len, name->name, dir->i_ino, dir->i_generation, dir);
990 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
991 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
996 /* Try to find the child dentry by its name.
997 If found, put the result fid into @fid. */
998 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1001 struct dentry *parent, *child;
1003 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1004 child = d_lookup(parent, name);
1007 *fid = *ll_inode2fid(child->d_inode);
1012 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1013 struct dentry *dchild, struct qstr *name)
1015 struct ptlrpc_request *request = NULL;
1016 struct md_op_data *op_data;
1020 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1021 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1023 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1026 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1027 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1028 if (IS_ERR(op_data))
1029 RETURN(PTR_ERR(op_data));
1031 ll_get_child_fid(dir, name, &op_data->op_fid3);
1032 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1033 ll_finish_md_op_data(op_data);
1035 ll_update_times(request, dir);
1036 ptlrpc_req_finished(request);
1040 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1042 struct mdt_body *body;
1043 struct lov_mds_md *eadata;
1044 struct lov_stripe_md *lsm = NULL;
1045 struct obd_trans_info oti = { 0 };
1047 struct obd_capa *oc = NULL;
1051 /* req is swabbed so this is safe */
1052 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1053 if (!(body->valid & OBD_MD_FLEASIZE))
1056 if (body->eadatasize == 0) {
1057 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1058 GOTO(out, rc = -EPROTO);
1061 /* The MDS sent back the EA because we unlinked the last reference
1062 * to this file. Use this EA to unlink the objects on the OST.
1063 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1064 * check it is complete and sensible. */
1065 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1067 LASSERT(eadata != NULL);
1069 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1071 CERROR("obd_unpackmd: %d\n", rc);
1074 LASSERT(rc >= sizeof(*lsm));
1078 GOTO(out_free_memmd, rc = -ENOMEM);
1080 oa->o_id = lsm->lsm_object_id;
1081 oa->o_seq = lsm->lsm_object_seq;
1082 oa->o_mode = body->mode & S_IFMT;
1083 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1085 if (body->valid & OBD_MD_FLCOOKIE) {
1086 oa->o_valid |= OBD_MD_FLCOOKIE;
1087 oti.oti_logcookies =
1088 req_capsule_server_sized_get(&request->rq_pill,
1090 sizeof(struct llog_cookie) *
1091 lsm->lsm_stripe_count);
1092 if (oti.oti_logcookies == NULL) {
1093 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1094 body->valid &= ~OBD_MD_FLCOOKIE;
1098 if (body->valid & OBD_MD_FLOSSCAPA) {
1099 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1101 GOTO(out_free_memmd, rc);
1104 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1108 CERROR("obd destroy objid "LPX64" error %d\n",
1109 lsm->lsm_object_id, rc);
1111 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1116 /* ll_unlink_generic() doesn't update the inode with the new link count.
1117 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1118 * is any lock existing. They will recycle dentries and inodes based upon locks
1120 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1121 struct dentry *dchild, struct qstr *name)
1123 struct ptlrpc_request *request = NULL;
1124 struct md_op_data *op_data;
1127 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1128 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1131 * XXX: unlink bind mountpoint maybe call to here,
1132 * just check it as vfs_unlink does.
1134 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1137 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1138 name->len, 0, LUSTRE_OPC_ANY, NULL);
1139 if (IS_ERR(op_data))
1140 RETURN(PTR_ERR(op_data));
1142 ll_get_child_fid(dir, name, &op_data->op_fid3);
1143 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1144 ll_finish_md_op_data(op_data);
1148 ll_update_times(request, dir);
1150 rc = ll_objects_destroy(request, dir);
1152 ptlrpc_req_finished(request);
1156 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1157 struct dentry *src_dchild, struct qstr *src_name,
1158 struct inode *tgt, struct dentry *tgt_dparent,
1159 struct dentry *tgt_dchild, struct qstr *tgt_name)
1161 struct ptlrpc_request *request = NULL;
1162 struct ll_sb_info *sbi = ll_i2sbi(src);
1163 struct md_op_data *op_data;
1166 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1167 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1168 src->i_ino, src->i_generation, src, tgt_name->len,
1169 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1171 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1172 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1175 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1176 LUSTRE_OPC_ANY, NULL);
1177 if (IS_ERR(op_data))
1178 RETURN(PTR_ERR(op_data));
1180 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1181 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1182 err = md_rename(sbi->ll_md_exp, op_data,
1183 src_name->name, src_name->len,
1184 tgt_name->name, tgt_name->len, &request);
1185 ll_finish_md_op_data(op_data);
1187 ll_update_times(request, src);
1188 ll_update_times(request, tgt);
1189 err = ll_objects_destroy(request, src);
1192 ptlrpc_req_finished(request);
1197 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1200 return ll_mknod_generic(dir, &dchild->d_name, mode,
1201 old_encode_dev(rdev), dchild);
1204 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1206 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1208 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1210 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1212 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1214 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1216 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1217 const char *oldname)
1219 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1221 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1222 struct dentry *new_dentry)
1224 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1227 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1228 struct inode *new_dir, struct dentry *new_dentry)
1231 err = ll_rename_generic(old_dir, NULL,
1232 old_dentry, &old_dentry->d_name,
1233 new_dir, NULL, new_dentry,
1234 &new_dentry->d_name);
1236 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1237 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1239 d_move(old_dentry, new_dentry);
1244 struct inode_operations ll_dir_inode_operations = {
1246 .lookup = ll_lookup_nd,
1247 .create = ll_create_nd,
1248 /* We need all these non-raw things for NFSD, to not patch it. */
1249 .unlink = ll_unlink,
1252 .symlink = ll_symlink,
1254 .rename = ll_rename,
1255 .setattr = ll_setattr,
1256 .getattr = ll_getattr,
1257 .permission = ll_inode_permission,
1258 .setxattr = ll_setxattr,
1259 .getxattr = ll_getxattr,
1260 .listxattr = ll_listxattr,
1261 .removexattr = ll_removexattr,
1264 struct inode_operations ll_special_inode_operations = {
1265 .setattr = ll_setattr,
1266 .getattr = ll_getattr,
1267 .permission = ll_inode_permission,
1268 .setxattr = ll_setxattr,
1269 .getxattr = ll_getxattr,
1270 .listxattr = ll_listxattr,
1271 .removexattr = ll_removexattr,