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 if (S_ISREG(inode->i_mode) &&
139 ll_i2info(inode)->lli_clob == NULL)
140 rc = cl_inode_init(inode, md);
143 make_bad_inode(inode);
144 unlock_new_inode(inode);
148 unlock_new_inode(inode);
149 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
150 ll_update_inode(inode, md);
151 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
152 inode, PFID(&md->body->fid1));
157 static void ll_drop_negative_dentry(struct inode *dir)
159 struct dentry *dentry, *tmp_alias, *tmp_subdir;
161 cfs_spin_lock(&ll_lookup_lock);
162 spin_lock(&dcache_lock);
164 list_for_each_entry_safe(dentry, tmp_alias,
165 &dir->i_dentry,d_alias) {
166 if (!list_empty(&dentry->d_subdirs)) {
167 struct dentry *child;
168 list_for_each_entry_safe(child, tmp_subdir,
171 /* XXX Print some debug here? */
173 /* Negative dentry. If we were
174 dropping dcache lock, go
175 throught the list again */
176 if (ll_drop_dentry(child))
181 spin_unlock(&dcache_lock);
182 cfs_spin_unlock(&ll_lookup_lock);
186 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
187 void *data, int flag)
190 struct lustre_handle lockh;
194 case LDLM_CB_BLOCKING:
195 ldlm_lock2handle(lock, &lockh);
196 rc = ldlm_cli_cancel(&lockh);
198 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
202 case LDLM_CB_CANCELING: {
203 struct inode *inode = ll_inode_from_lock(lock);
204 struct ll_inode_info *lli;
205 __u64 bits = lock->l_policy_data.l_inodebits.bits;
207 ldlm_mode_t mode = lock->l_req_mode;
209 /* Invalidate all dentries associated with this inode */
213 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
214 /* For OPEN locks we differentiate between lock modes - CR, CW. PR - bug 22891 */
215 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE))
216 ll_have_md_lock(inode, &bits, LCK_MINMODE);
218 if (bits & MDS_INODELOCK_OPEN)
219 ll_have_md_lock(inode, &bits, mode);
221 fid = ll_inode2fid(inode);
222 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
223 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
224 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
225 LDLM_ERROR(lock, "data mismatch with object "
226 DFID" (%p)", PFID(fid), inode);
229 if (bits & MDS_INODELOCK_OPEN) {
231 switch (lock->l_req_mode) {
242 CERROR("Unexpected lock mode for OPEN lock "
243 "%d, inode %ld\n", lock->l_req_mode,
246 ll_md_real_close(inode, flags);
249 lli = ll_i2info(inode);
250 if (bits & MDS_INODELOCK_UPDATE)
251 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
253 if (S_ISDIR(inode->i_mode) &&
254 (bits & MDS_INODELOCK_UPDATE)) {
255 CDEBUG(D_INODE, "invalidating inode %lu\n",
257 truncate_inode_pages(inode->i_mapping, 0);
258 ll_drop_negative_dentry(inode);
261 if (inode->i_sb->s_root &&
262 inode != inode->i_sb->s_root->d_inode &&
263 (bits & MDS_INODELOCK_LOOKUP))
264 ll_unhash_aliases(inode);
275 __u32 ll_i2suppgid(struct inode *i)
277 if (cfs_curproc_is_in_groups(i->i_gid))
278 return (__u32)i->i_gid;
283 /* Pack the required supplementary groups into the supplied groups array.
284 * If we don't need to use the groups from the target inode(s) then we
285 * instead pack one or more groups from the user's supplementary group
286 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
287 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
294 LASSERT(suppgids != NULL);
296 suppgids[0] = ll_i2suppgid(i1);
299 suppgids[1] = ll_i2suppgid(i2);
304 for (i = 0; i < current_ngroups; i++) {
305 if (suppgids[0] == -1) {
306 if (current_groups[i] != suppgids[1])
307 suppgids[0] = current_groups[i];
310 if (suppgids[1] == -1) {
311 if (current_groups[i] != suppgids[0])
312 suppgids[1] = current_groups[i];
320 static void ll_d_add(struct dentry *de, struct inode *inode)
322 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
324 if (!list_empty(&de->d_alias)) {
325 spin_unlock(&dcache_lock);
326 CERROR("dentry %.*s %p alias next %p, prev %p\n",
327 de->d_name.len, de->d_name.name, de,
328 de->d_alias.next, de->d_alias.prev);
332 list_add(&de->d_alias, &inode->i_dentry);
336 if (!d_unhashed(de)) {
337 spin_unlock(&dcache_lock);
338 CERROR("dentry %.*s %p hash next %p\n",
339 de->d_name.len, de->d_name.name, de, de->d_hash.next);
342 d_rehash_cond(de, 0);
345 /* Search "inode"'s alias list for a dentry that has the same name and parent
346 * as de. If found, return it. If not found, return de.
347 * Lustre can't use d_add_unique because don't unhash aliases for directory
348 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
349 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
351 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
353 struct list_head *tmp;
354 struct dentry *dentry;
355 struct dentry *last_discon = NULL;
357 cfs_spin_lock(&ll_lookup_lock);
358 spin_lock(&dcache_lock);
359 list_for_each(tmp, &inode->i_dentry) {
360 dentry = list_entry(tmp, struct dentry, d_alias);
362 /* We are called here with 'de' already on the aliases list. */
363 if (unlikely(dentry == de)) {
368 if (dentry->d_flags & DCACHE_DISCONNECTED) {
369 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
370 last_discon = dentry;
374 if (dentry->d_parent != de->d_parent)
377 if (dentry->d_name.hash != de->d_name.hash)
380 if (dentry->d_name.len != de->d_name.len)
383 if (memcmp(dentry->d_name.name, de->d_name.name,
384 de->d_name.len) != 0)
388 ll_dops_init(dentry, 0, 1);
389 ll_dentry_rehash(dentry, 1);
390 spin_unlock(&dcache_lock);
391 cfs_spin_unlock(&ll_lookup_lock);
393 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
394 "refc %d\n", de->d_name.len, de->d_name.name, de,
395 de->d_parent, de->d_inode, atomic_read(&de->d_count));
400 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
401 "refc %d\n", last_discon, last_discon->d_inode,
402 atomic_read(&last_discon->d_count));
403 dget_locked(last_discon);
404 lock_dentry(last_discon);
405 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
406 unlock_dentry(last_discon);
407 spin_unlock(&dcache_lock);
408 cfs_spin_unlock(&ll_lookup_lock);
409 ll_dops_init(last_discon, 1, 1);
411 d_move(last_discon, de);
416 de->d_flags |= DCACHE_LUSTRE_INVALID;
420 spin_unlock(&dcache_lock);
421 cfs_spin_unlock(&ll_lookup_lock);
426 int ll_lookup_it_finish(struct ptlrpc_request *request,
427 struct lookup_intent *it, void *data)
429 struct it_cb_data *icbd = data;
430 struct dentry **de = icbd->icbd_childp;
431 struct inode *parent = icbd->icbd_parent;
432 struct inode *inode = NULL;
436 /* NB 1 request reference will be taken away by ll_intent_lock()
438 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
439 struct dentry *save = *de;
442 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
446 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
448 /* We used to query real size from OSTs here, but actually
449 this is not needed. For stat() calls size would be updated
450 from subsequent do_revalidate()->ll_inode_revalidate_it() in
452 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
453 Everybody else who needs correct file size would call
454 cl_glimpse_size or some equivalent themselves anyway.
455 Also see bug 7198. */
456 ll_dops_init(*de, 1, 1);
457 *de = ll_find_alias(inode, *de);
459 struct ll_dentry_data *lld = ll_d2d(*de);
461 /* just make sure the ll_dentry_data is ready */
462 if (unlikely(lld == NULL))
463 ll_dops_init(*de, 1, 1);
465 /* we have lookup look - unhide dentry */
466 ll_dentry_reset_flags(*de, bits);
470 ll_dops_init(*de, 1, 1);
471 /* Check that parent has UPDATE lock. If there is none, we
472 cannot afford to hash this dentry (done by ll_d_add) as it
473 might get picked up later when UPDATE lock will appear */
474 ibits = MDS_INODELOCK_UPDATE;
475 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE)) {
476 spin_lock(&dcache_lock);
478 spin_unlock(&dcache_lock);
480 /* negative lookup - and don't have update lock to
483 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
486 (*de)->d_inode = NULL;
487 /* We do not want to hash the dentry if don`t have a
488 * lock, but if this dentry is later used in d_move,
489 * we'd hit uninitialised list head d_hash, so we just
490 * do this to init d_hash field but leave dentry
491 * unhashed. (bug 10796). */
500 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
501 struct lookup_intent *it, int lookup_flags)
503 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
504 struct dentry *save = dentry, *retval;
505 struct ptlrpc_request *req = NULL;
506 struct md_op_data *op_data;
507 struct it_cb_data icbd;
512 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
513 RETURN(ERR_PTR(-ENAMETOOLONG));
515 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
516 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
517 parent->i_generation, parent, LL_IT2STR(it));
519 if (d_mountpoint(dentry))
520 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
522 ll_frob_intent(&it, &lookup_it);
524 /* As do_lookup is called before follow_mount, root dentry may be left
525 * not valid, revalidate it here. */
526 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
527 (it->it_op & (IT_OPEN | IT_CREAT))) {
528 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
529 MDS_INODELOCK_LOOKUP);
534 if (it->it_op == IT_GETATTR) {
535 rc = ll_statahead_enter(parent, &dentry, 0);
538 GOTO(out, retval = NULL);
539 GOTO(out, retval = dentry);
543 icbd.icbd_childp = &dentry;
544 icbd.icbd_parent = parent;
546 if (it->it_op & IT_CREAT ||
547 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
548 opc = LUSTRE_OPC_CREATE;
550 opc = LUSTRE_OPC_ANY;
552 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
553 dentry->d_name.len, lookup_flags, opc,
556 RETURN((void *)op_data);
558 it->it_create_mode &= ~cfs_curproc_umask();
560 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
561 lookup_flags, &req, ll_md_blocking_ast, 0);
562 ll_finish_md_op_data(op_data);
564 GOTO(out, retval = ERR_PTR(rc));
566 rc = ll_lookup_it_finish(req, it, &icbd);
568 ll_intent_release(it);
569 GOTO(out, retval = ERR_PTR(rc));
572 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
573 !S_ISREG(dentry->d_inode->i_mode) &&
574 !S_ISDIR(dentry->d_inode->i_mode)) {
575 ll_release_openhandle(dentry, it);
577 ll_lookup_finish_locks(it, dentry);
580 GOTO(out, retval = NULL);
582 GOTO(out, retval = dentry);
585 ptlrpc_req_finished(req);
586 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
587 ll_statahead_mark(parent, dentry);
591 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
594 struct lookup_intent *it;
596 OBD_ALLOC(it, sizeof(*it));
598 return ERR_PTR(-ENOMEM);
600 if (lookup_flags & LOOKUP_OPEN) {
602 if (lookup_flags & LOOKUP_CREATE)
603 it->it_op |= IT_CREAT;
604 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
605 it->it_flags = oit->flags;
607 it->it_op = IT_GETATTR;
610 #ifndef HAVE_FILE_IN_STRUCT_INTENT
611 /* Since there is no way to pass our intent to ll_file_open,
612 * just check the file is there. Actual open will be done
614 if (it->it_op & IT_OPEN)
615 it->it_op = IT_LOOKUP;
621 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
622 struct nameidata *nd)
627 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
628 struct lookup_intent *it;
630 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
631 /* Did we came here from failed revalidate just to propagate
633 if (nd->flags & LOOKUP_OPEN)
634 if (IS_ERR(nd->intent.open.file))
635 RETURN((struct dentry *)nd->intent.open.file);
637 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
638 it = ll_d2d(dentry)->lld_it;
639 ll_d2d(dentry)->lld_it = NULL;
641 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
642 /* We are sure this is new dentry, so we need to create
643 our private data and set the dentry ops */
644 ll_dops_init(dentry, 1, 1);
647 it = ll_convert_intent(&nd->intent.open, nd->flags);
649 RETURN((struct dentry *)it);
652 de = ll_lookup_it(parent, dentry, it, nd->flags);
655 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
656 if (dentry->d_inode &&
657 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
658 #ifdef HAVE_FILE_IN_STRUCT_INTENT
659 if (S_ISFIFO(dentry->d_inode->i_mode)) {
660 // We cannot call open here as it would
663 (struct ptlrpc_request *)
664 it->d.lustre.it_data);
666 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
667 /* 2.6.1[456] have a bug in open_namei() that forgets to check
668 * nd->intent.open.file for error, so we need to return it as lookup's result
671 nd->intent.open.file->private_data = it;
672 filp =lookup_instantiate_filp(nd,dentry,
677 de = (struct dentry *) filp;
680 nd->intent.open.file->private_data = it;
681 (void)lookup_instantiate_filp(nd,dentry,
686 #else /* HAVE_FILE_IN_STRUCT_INTENT */
687 /* Release open handle as we have no way to
688 * pass it to ll_file_open */
689 ll_release_openhandle(dentry, it);
690 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
691 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
692 // XXX This can only reliably work on assumption
693 // that there are NO hashed negative dentries.
694 ll_d2d(dentry)->lld_it = it;
695 it = NULL; /* Will be freed in ll_create_nd */
696 /* We absolutely depend on ll_create_nd to be
697 * called to not leak this intent and possible
698 * data attached to it */
703 ll_intent_release(it);
704 OBD_FREE(it, sizeof(*it));
707 de = ll_lookup_it(parent, dentry, NULL, 0);
713 /* We depend on "mode" being set with the proper file type/umask by now */
714 static struct inode *ll_create_node(struct inode *dir, const char *name,
715 int namelen, const void *data, int datalen,
716 int mode, __u64 extra,
717 struct lookup_intent *it)
719 struct inode *inode = NULL;
720 struct ptlrpc_request *request = NULL;
721 struct ll_sb_info *sbi = ll_i2sbi(dir);
725 LASSERT(it && it->d.lustre.it_disposition);
727 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
728 request = it->d.lustre.it_data;
729 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
730 rc = ll_prep_inode(&inode, request, dir->i_sb);
732 GOTO(out, inode = ERR_PTR(rc));
734 LASSERT(list_empty(&inode->i_dentry));
736 /* We asked for a lock on the directory, but were granted a
737 * lock on the inode. Since we finally have an inode pointer,
738 * stuff it in the lock. */
739 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
740 inode, inode->i_ino, inode->i_generation);
741 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
744 ptlrpc_req_finished(request);
749 * By the time this is called, we already have created the directory cache
750 * entry for the new file, but it is so far negative - it has no inode.
752 * We defer creating the OBD object(s) until open, to keep the intent and
753 * non-intent code paths similar, and also because we do not have the MDS
754 * inode number before calling ll_create_node() (which is needed for LOV),
755 * so we would need to do yet another RPC to the MDS to store the LOV EA
756 * data on the MDS. If needed, we would pass the PACKED lmm as data and
757 * lmm_size in datalen (the MDS still has code which will handle that).
759 * If the create succeeds, we fill in the inode information
760 * with d_instantiate().
762 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
763 struct lookup_intent *it)
769 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
770 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
771 dir->i_generation, dir, LL_IT2STR(it));
773 rc = it_open_error(DISP_OPEN_CREATE, it);
777 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
778 NULL, 0, mode, 0, it);
780 RETURN(PTR_ERR(inode));
782 d_instantiate(dentry, inode);
783 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
784 * but some callers, e.g. do_coredump, expect dentry to be hashed after
785 * successful create. Hash it here. */
786 spin_lock(&dcache_lock);
787 if (d_unhashed(dentry))
788 d_rehash_cond(dentry, 0);
789 spin_unlock(&dcache_lock);
793 static void ll_update_times(struct ptlrpc_request *request,
796 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
800 if (body->valid & OBD_MD_FLMTIME &&
801 body->mtime > LTIME_S(inode->i_mtime)) {
802 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
803 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
804 LTIME_S(inode->i_mtime) = body->mtime;
806 if (body->valid & OBD_MD_FLCTIME &&
807 body->ctime > LTIME_S(inode->i_ctime))
808 LTIME_S(inode->i_ctime) = body->ctime;
811 static int ll_new_node(struct inode *dir, struct qstr *name,
812 const char *tgt, int mode, int rdev,
813 struct dentry *dchild, __u32 opc)
815 struct ptlrpc_request *request = NULL;
816 struct md_op_data *op_data;
817 struct inode *inode = NULL;
818 struct ll_sb_info *sbi = ll_i2sbi(dir);
823 if (unlikely(tgt != NULL))
824 tgt_len = strlen(tgt) + 1;
826 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
827 name->len, 0, opc, NULL);
829 GOTO(err_exit, err = PTR_ERR(op_data));
831 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
832 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
833 cfs_curproc_cap_pack(), rdev, &request);
834 ll_finish_md_op_data(op_data);
838 ll_update_times(request, dir);
841 err = ll_prep_inode(&inode, request, dchild->d_sb);
846 d_instantiate(dchild, inode);
850 ptlrpc_req_finished(request);
855 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
856 unsigned rdev, struct dentry *dchild)
861 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
862 name->len, name->name, dir->i_ino, dir->i_generation, dir,
865 mode &= ~cfs_curproc_umask();
867 switch (mode & S_IFMT) {
869 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
875 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
887 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
888 int mode, struct nameidata *nd)
890 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
894 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
896 ll_d2d(dentry)->lld_it = NULL;
898 /* Was there an error? Propagate it! */
899 if (it->d.lustre.it_status) {
900 rc = it->d.lustre.it_status;
904 rc = ll_create_it(dir, dentry, mode, it);
905 #ifdef HAVE_FILE_IN_STRUCT_INTENT
906 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
907 nd->intent.open.file->private_data = it;
908 lookup_instantiate_filp(nd, dentry, NULL);
911 ll_release_openhandle(dentry,it);
915 ll_intent_release(it);
916 OBD_FREE(it, sizeof(*it));
921 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
922 const char *tgt, struct dentry *dchild)
927 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
928 name->len, name->name, dir->i_ino, dir->i_generation,
931 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
932 0, dchild, LUSTRE_OPC_SYMLINK);
936 static int ll_link_generic(struct inode *src, struct inode *dir,
937 struct qstr *name, struct dentry *dchild)
939 struct ll_sb_info *sbi = ll_i2sbi(dir);
940 struct ptlrpc_request *request = NULL;
941 struct md_op_data *op_data;
946 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
947 src->i_ino, src->i_generation, src, dir->i_ino,
948 dir->i_generation, dir, name->len, name->name);
950 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
951 0, LUSTRE_OPC_ANY, NULL);
953 RETURN(PTR_ERR(op_data));
955 err = md_link(sbi->ll_md_exp, op_data, &request);
956 ll_finish_md_op_data(op_data);
962 ll_update_times(request, dir);
965 ptlrpc_req_finished(request);
969 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
970 int mode, struct dentry *dchild)
976 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
977 name->len, name->name, dir->i_ino, dir->i_generation, dir);
979 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
980 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
985 /* Try to find the child dentry by its name.
986 If found, put the result fid into @fid. */
987 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
990 struct dentry *parent, *child;
992 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
993 child = d_lookup(parent, name);
996 *fid = *ll_inode2fid(child->d_inode);
1001 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1002 struct dentry *dchild, struct qstr *name)
1004 struct ptlrpc_request *request = NULL;
1005 struct md_op_data *op_data;
1009 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1010 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1012 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1015 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1016 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1017 if (IS_ERR(op_data))
1018 RETURN(PTR_ERR(op_data));
1020 ll_get_child_fid(dir, name, &op_data->op_fid3);
1021 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1022 ll_finish_md_op_data(op_data);
1024 ll_update_times(request, dir);
1025 ptlrpc_req_finished(request);
1029 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1031 struct mdt_body *body;
1032 struct lov_mds_md *eadata;
1033 struct lov_stripe_md *lsm = NULL;
1034 struct obd_trans_info oti = { 0 };
1036 struct obd_capa *oc = NULL;
1040 /* req is swabbed so this is safe */
1041 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1042 if (!(body->valid & OBD_MD_FLEASIZE))
1045 if (body->eadatasize == 0) {
1046 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1047 GOTO(out, rc = -EPROTO);
1050 /* The MDS sent back the EA because we unlinked the last reference
1051 * to this file. Use this EA to unlink the objects on the OST.
1052 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1053 * check it is complete and sensible. */
1054 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1056 LASSERT(eadata != NULL);
1058 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1060 CERROR("obd_unpackmd: %d\n", rc);
1063 LASSERT(rc >= sizeof(*lsm));
1067 GOTO(out_free_memmd, rc = -ENOMEM);
1069 oa->o_id = lsm->lsm_object_id;
1070 oa->o_seq = lsm->lsm_object_seq;
1071 oa->o_mode = body->mode & S_IFMT;
1072 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1074 if (body->valid & OBD_MD_FLCOOKIE) {
1075 oa->o_valid |= OBD_MD_FLCOOKIE;
1076 oti.oti_logcookies =
1077 req_capsule_server_sized_get(&request->rq_pill,
1079 sizeof(struct llog_cookie) *
1080 lsm->lsm_stripe_count);
1081 if (oti.oti_logcookies == NULL) {
1082 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1083 body->valid &= ~OBD_MD_FLCOOKIE;
1087 if (body->valid & OBD_MD_FLOSSCAPA) {
1088 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1090 GOTO(out_free_memmd, rc);
1093 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1097 CERROR("obd destroy objid "LPX64" error %d\n",
1098 lsm->lsm_object_id, rc);
1100 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1105 /* ll_unlink_generic() doesn't update the inode with the new link count.
1106 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1107 * is any lock existing. They will recycle dentries and inodes based upon locks
1109 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1110 struct dentry *dchild, struct qstr *name)
1112 struct ptlrpc_request *request = NULL;
1113 struct md_op_data *op_data;
1116 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1117 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1120 * XXX: unlink bind mountpoint maybe call to here,
1121 * just check it as vfs_unlink does.
1123 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1126 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1127 name->len, 0, 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 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1133 ll_finish_md_op_data(op_data);
1137 ll_update_times(request, dir);
1139 rc = ll_objects_destroy(request, dir);
1141 ptlrpc_req_finished(request);
1145 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1146 struct dentry *src_dchild, struct qstr *src_name,
1147 struct inode *tgt, struct dentry *tgt_dparent,
1148 struct dentry *tgt_dchild, struct qstr *tgt_name)
1150 struct ptlrpc_request *request = NULL;
1151 struct ll_sb_info *sbi = ll_i2sbi(src);
1152 struct md_op_data *op_data;
1155 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1156 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1157 src->i_ino, src->i_generation, src, tgt_name->len,
1158 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1160 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1161 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1164 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1165 LUSTRE_OPC_ANY, NULL);
1166 if (IS_ERR(op_data))
1167 RETURN(PTR_ERR(op_data));
1169 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1170 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1171 err = md_rename(sbi->ll_md_exp, op_data,
1172 src_name->name, src_name->len,
1173 tgt_name->name, tgt_name->len, &request);
1174 ll_finish_md_op_data(op_data);
1176 ll_update_times(request, src);
1177 ll_update_times(request, tgt);
1178 err = ll_objects_destroy(request, src);
1181 ptlrpc_req_finished(request);
1186 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1189 return ll_mknod_generic(dir, &dchild->d_name, mode,
1190 old_encode_dev(rdev), dchild);
1193 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1195 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1197 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1199 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1201 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1203 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1205 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1206 const char *oldname)
1208 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1210 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1211 struct dentry *new_dentry)
1213 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1216 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1217 struct inode *new_dir, struct dentry *new_dentry)
1220 err = ll_rename_generic(old_dir, NULL,
1221 old_dentry, &old_dentry->d_name,
1222 new_dir, NULL, new_dentry,
1223 &new_dentry->d_name);
1225 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1226 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1228 d_move(old_dentry, new_dentry);
1233 struct inode_operations ll_dir_inode_operations = {
1235 .lookup = ll_lookup_nd,
1236 .create = ll_create_nd,
1237 /* We need all these non-raw things for NFSD, to not patch it. */
1238 .unlink = ll_unlink,
1241 .symlink = ll_symlink,
1243 .rename = ll_rename,
1244 .setattr = ll_setattr,
1245 .getattr = ll_getattr,
1246 .permission = ll_inode_permission,
1247 .setxattr = ll_setxattr,
1248 .getxattr = ll_getxattr,
1249 .listxattr = ll_listxattr,
1250 .removexattr = ll_removexattr,
1253 struct inode_operations ll_special_inode_operations = {
1254 .setattr = ll_setattr,
1255 .getattr = ll_getattr,
1256 .permission = ll_inode_permission,
1257 .setxattr = ll_setxattr,
1258 .getxattr = ll_getxattr,
1259 .listxattr = ll_listxattr,
1260 .removexattr = ll_removexattr,