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 static 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)
390 unlock_dentry(dentry);
391 ll_dops_init(dentry, 0, 1);
392 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
393 spin_unlock(&dcache_lock);
394 cfs_spin_unlock(&ll_lookup_lock);
396 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
397 "refc %d\n", de->d_name.len, de->d_name.name, de,
398 de->d_parent, de->d_inode, atomic_read(&de->d_count));
403 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
404 "refc %d\n", last_discon, last_discon->d_inode,
405 atomic_read(&last_discon->d_count));
406 dget_locked(last_discon);
407 lock_dentry(last_discon);
408 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
409 unlock_dentry(last_discon);
410 spin_unlock(&dcache_lock);
411 cfs_spin_unlock(&ll_lookup_lock);
412 ll_dops_init(last_discon, 1, 1);
414 d_move(last_discon, de);
419 de->d_flags |= DCACHE_LUSTRE_INVALID;
423 spin_unlock(&dcache_lock);
424 cfs_spin_unlock(&ll_lookup_lock);
429 int ll_lookup_it_finish(struct ptlrpc_request *request,
430 struct lookup_intent *it, void *data)
432 struct it_cb_data *icbd = data;
433 struct dentry **de = icbd->icbd_childp;
434 struct inode *parent = icbd->icbd_parent;
435 struct inode *inode = NULL;
439 /* NB 1 request reference will be taken away by ll_intent_lock()
441 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
442 struct dentry *save = *de;
445 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
449 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
451 /* We used to query real size from OSTs here, but actually
452 this is not needed. For stat() calls size would be updated
453 from subsequent do_revalidate()->ll_inode_revalidate_it() in
455 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
456 Everybody else who needs correct file size would call
457 cl_glimpse_size or some equivalent themselves anyway.
458 Also see bug 7198. */
459 ll_dops_init(*de, 1, 1);
460 *de = ll_find_alias(inode, *de);
462 struct ll_dentry_data *lld = ll_d2d(*de);
464 /* just make sure the ll_dentry_data is ready */
465 if (unlikely(lld == NULL))
466 ll_dops_init(*de, 1, 1);
468 /* we have lookup look - unhide dentry */
469 if ((*de)->d_flags & DCACHE_LUSTRE_INVALID &&
470 bits & MDS_INODELOCK_LOOKUP) {
472 (*de)->d_flags &= ~DCACHE_LUSTRE_INVALID;
478 ll_dops_init(*de, 1, 1);
479 /* Check that parent has UPDATE lock. If there is none, we
480 cannot afford to hash this dentry (done by ll_d_add) as it
481 might get picked up later when UPDATE lock will appear */
482 ibits = MDS_INODELOCK_UPDATE;
483 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE)) {
484 spin_lock(&dcache_lock);
486 spin_unlock(&dcache_lock);
488 /* negative lookup - and don't have update lock to
491 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
494 (*de)->d_inode = NULL;
495 /* We do not want to hash the dentry if don`t have a
496 * lock, but if this dentry is later used in d_move,
497 * we'd hit uninitialised list head d_hash, so we just
498 * do this to init d_hash field but leave dentry
499 * unhashed. (bug 10796). */
508 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
509 struct lookup_intent *it, int lookup_flags)
511 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
512 struct dentry *save = dentry, *retval;
513 struct ptlrpc_request *req = NULL;
514 struct md_op_data *op_data;
515 struct it_cb_data icbd;
520 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
521 RETURN(ERR_PTR(-ENAMETOOLONG));
523 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
524 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
525 parent->i_generation, parent, LL_IT2STR(it));
527 if (d_mountpoint(dentry))
528 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
530 ll_frob_intent(&it, &lookup_it);
532 /* As do_lookup is called before follow_mount, root dentry may be left
533 * not valid, revalidate it here. */
534 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
535 (it->it_op & (IT_OPEN | IT_CREAT))) {
536 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
537 MDS_INODELOCK_LOOKUP);
542 if (it->it_op == IT_GETATTR) {
543 first = ll_statahead_enter(parent, &dentry, 1);
545 ll_statahead_exit(parent, dentry, first);
547 RETURN(retval = dentry);
551 icbd.icbd_childp = &dentry;
552 icbd.icbd_parent = parent;
554 if (it->it_op & IT_CREAT ||
555 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
556 opc = LUSTRE_OPC_CREATE;
558 opc = LUSTRE_OPC_ANY;
560 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
561 dentry->d_name.len, lookup_flags, opc,
564 RETURN((void *)op_data);
566 it->it_create_mode &= ~cfs_curproc_umask();
568 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
569 lookup_flags, &req, ll_md_blocking_ast, 0);
570 ll_finish_md_op_data(op_data);
572 GOTO(out, retval = ERR_PTR(rc));
574 rc = ll_lookup_it_finish(req, it, &icbd);
576 ll_intent_release(it);
577 GOTO(out, retval = ERR_PTR(rc));
580 if (first == -EEXIST)
581 ll_statahead_mark(parent, dentry);
583 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
584 !S_ISREG(dentry->d_inode->i_mode) &&
585 !S_ISDIR(dentry->d_inode->i_mode)) {
586 ll_release_openhandle(dentry, it);
588 ll_lookup_finish_locks(it, dentry);
591 GOTO(out, retval = NULL);
593 GOTO(out, retval = dentry);
596 ptlrpc_req_finished(req);
600 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
603 struct lookup_intent *it;
605 OBD_ALLOC(it, sizeof(*it));
607 return ERR_PTR(-ENOMEM);
609 if (lookup_flags & LOOKUP_OPEN) {
611 if (lookup_flags & LOOKUP_CREATE)
612 it->it_op |= IT_CREAT;
613 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
614 it->it_flags = oit->flags;
616 it->it_op = IT_GETATTR;
619 #ifndef HAVE_FILE_IN_STRUCT_INTENT
620 /* Since there is no way to pass our intent to ll_file_open,
621 * just check the file is there. Actual open will be done
623 if (it->it_op & IT_OPEN)
624 it->it_op = IT_LOOKUP;
630 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
631 struct nameidata *nd)
636 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
637 struct lookup_intent *it;
639 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
640 /* Did we came here from failed revalidate just to propagate
642 if (nd->flags & LOOKUP_OPEN)
643 if (IS_ERR(nd->intent.open.file))
644 RETURN((struct dentry *)nd->intent.open.file);
646 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
647 it = ll_d2d(dentry)->lld_it;
648 ll_d2d(dentry)->lld_it = NULL;
650 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
651 /* We are sure this is new dentry, so we need to create
652 our private data and set the dentry ops */
653 ll_dops_init(dentry, 1, 1);
656 it = ll_convert_intent(&nd->intent.open, nd->flags);
658 RETURN((struct dentry *)it);
661 de = ll_lookup_it(parent, dentry, it, nd->flags);
664 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
665 if (dentry->d_inode &&
666 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
667 #ifdef HAVE_FILE_IN_STRUCT_INTENT
668 if (S_ISFIFO(dentry->d_inode->i_mode)) {
669 // We cannot call open here as it would
672 (struct ptlrpc_request *)
673 it->d.lustre.it_data);
675 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
676 /* 2.6.1[456] have a bug in open_namei() that forgets to check
677 * nd->intent.open.file for error, so we need to return it as lookup's result
680 nd->intent.open.file->private_data = it;
681 filp =lookup_instantiate_filp(nd,dentry,
686 de = (struct dentry *) filp;
689 nd->intent.open.file->private_data = it;
690 (void)lookup_instantiate_filp(nd,dentry,
695 #else /* HAVE_FILE_IN_STRUCT_INTENT */
696 /* Release open handle as we have no way to
697 * pass it to ll_file_open */
698 ll_release_openhandle(dentry, it);
699 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
700 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
701 // XXX This can only reliably work on assumption
702 // that there are NO hashed negative dentries.
703 ll_d2d(dentry)->lld_it = it;
704 it = NULL; /* Will be freed in ll_create_nd */
705 /* We absolutely depend on ll_create_nd to be
706 * called to not leak this intent and possible
707 * data attached to it */
712 ll_intent_release(it);
713 OBD_FREE(it, sizeof(*it));
716 de = ll_lookup_it(parent, dentry, NULL, 0);
722 /* We depend on "mode" being set with the proper file type/umask by now */
723 static struct inode *ll_create_node(struct inode *dir, const char *name,
724 int namelen, const void *data, int datalen,
725 int mode, __u64 extra,
726 struct lookup_intent *it)
728 struct inode *inode = NULL;
729 struct ptlrpc_request *request = NULL;
730 struct ll_sb_info *sbi = ll_i2sbi(dir);
734 LASSERT(it && it->d.lustre.it_disposition);
736 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
737 request = it->d.lustre.it_data;
738 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
739 rc = ll_prep_inode(&inode, request, dir->i_sb);
741 GOTO(out, inode = ERR_PTR(rc));
743 LASSERT(list_empty(&inode->i_dentry));
745 /* We asked for a lock on the directory, but were granted a
746 * lock on the inode. Since we finally have an inode pointer,
747 * stuff it in the lock. */
748 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
749 inode, inode->i_ino, inode->i_generation);
750 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
753 ptlrpc_req_finished(request);
758 * By the time this is called, we already have created the directory cache
759 * entry for the new file, but it is so far negative - it has no inode.
761 * We defer creating the OBD object(s) until open, to keep the intent and
762 * non-intent code paths similar, and also because we do not have the MDS
763 * inode number before calling ll_create_node() (which is needed for LOV),
764 * so we would need to do yet another RPC to the MDS to store the LOV EA
765 * data on the MDS. If needed, we would pass the PACKED lmm as data and
766 * lmm_size in datalen (the MDS still has code which will handle that).
768 * If the create succeeds, we fill in the inode information
769 * with d_instantiate().
771 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
772 struct lookup_intent *it)
778 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
779 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
780 dir->i_generation, dir, LL_IT2STR(it));
782 rc = it_open_error(DISP_OPEN_CREATE, it);
786 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
787 NULL, 0, mode, 0, it);
789 RETURN(PTR_ERR(inode));
791 d_instantiate(dentry, inode);
792 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
793 * but some callers, e.g. do_coredump, expect dentry to be hashed after
794 * successful create. Hash it here. */
795 spin_lock(&dcache_lock);
796 if (d_unhashed(dentry))
797 d_rehash_cond(dentry, 0);
798 spin_unlock(&dcache_lock);
802 static void ll_update_times(struct ptlrpc_request *request,
805 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
809 if (body->valid & OBD_MD_FLMTIME &&
810 body->mtime > LTIME_S(inode->i_mtime)) {
811 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
812 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
813 LTIME_S(inode->i_mtime) = body->mtime;
815 if (body->valid & OBD_MD_FLCTIME &&
816 body->ctime > LTIME_S(inode->i_ctime))
817 LTIME_S(inode->i_ctime) = body->ctime;
820 static int ll_new_node(struct inode *dir, struct qstr *name,
821 const char *tgt, int mode, int rdev,
822 struct dentry *dchild, __u32 opc)
824 struct ptlrpc_request *request = NULL;
825 struct md_op_data *op_data;
826 struct inode *inode = NULL;
827 struct ll_sb_info *sbi = ll_i2sbi(dir);
832 if (unlikely(tgt != NULL))
833 tgt_len = strlen(tgt) + 1;
835 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
836 name->len, 0, opc, NULL);
838 GOTO(err_exit, err = PTR_ERR(op_data));
840 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
841 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
842 cfs_curproc_cap_pack(), rdev, &request);
843 ll_finish_md_op_data(op_data);
847 ll_update_times(request, dir);
850 err = ll_prep_inode(&inode, request, dchild->d_sb);
855 d_instantiate(dchild, inode);
859 ptlrpc_req_finished(request);
864 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
865 unsigned rdev, struct dentry *dchild)
870 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
871 name->len, name->name, dir->i_ino, dir->i_generation, dir,
874 mode &= ~cfs_curproc_umask();
876 switch (mode & S_IFMT) {
878 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
884 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
896 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
897 int mode, struct nameidata *nd)
899 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
903 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
905 ll_d2d(dentry)->lld_it = NULL;
907 /* Was there an error? Propagate it! */
908 if (it->d.lustre.it_status) {
909 rc = it->d.lustre.it_status;
913 rc = ll_create_it(dir, dentry, mode, it);
914 #ifdef HAVE_FILE_IN_STRUCT_INTENT
915 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
916 nd->intent.open.file->private_data = it;
917 lookup_instantiate_filp(nd, dentry, NULL);
920 ll_release_openhandle(dentry,it);
924 ll_intent_release(it);
925 OBD_FREE(it, sizeof(*it));
930 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
931 const char *tgt, struct dentry *dchild)
936 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
937 name->len, name->name, dir->i_ino, dir->i_generation,
940 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
941 0, dchild, LUSTRE_OPC_SYMLINK);
945 static int ll_link_generic(struct inode *src, struct inode *dir,
946 struct qstr *name, struct dentry *dchild)
948 struct ll_sb_info *sbi = ll_i2sbi(dir);
949 struct ptlrpc_request *request = NULL;
950 struct md_op_data *op_data;
955 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
956 src->i_ino, src->i_generation, src, dir->i_ino,
957 dir->i_generation, dir, name->len, name->name);
959 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
960 0, LUSTRE_OPC_ANY, NULL);
962 RETURN(PTR_ERR(op_data));
964 err = md_link(sbi->ll_md_exp, op_data, &request);
965 ll_finish_md_op_data(op_data);
971 ll_update_times(request, dir);
974 ptlrpc_req_finished(request);
978 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
979 int mode, struct dentry *dchild)
985 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
986 name->len, name->name, dir->i_ino, dir->i_generation, dir);
988 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
989 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
994 /* Try to find the child dentry by its name.
995 If found, put the result fid into @fid. */
996 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
999 struct dentry *parent, *child;
1001 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1002 child = d_lookup(parent, name);
1005 *fid = *ll_inode2fid(child->d_inode);
1010 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1011 struct dentry *dchild, struct qstr *name)
1013 struct ptlrpc_request *request = NULL;
1014 struct md_op_data *op_data;
1018 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1019 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1021 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1024 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1025 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1026 if (IS_ERR(op_data))
1027 RETURN(PTR_ERR(op_data));
1029 ll_get_child_fid(dir, name, &op_data->op_fid3);
1030 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1031 ll_finish_md_op_data(op_data);
1033 ll_update_times(request, dir);
1034 ptlrpc_req_finished(request);
1038 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1040 struct mdt_body *body;
1041 struct lov_mds_md *eadata;
1042 struct lov_stripe_md *lsm = NULL;
1043 struct obd_trans_info oti = { 0 };
1045 struct obd_capa *oc = NULL;
1049 /* req is swabbed so this is safe */
1050 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1051 if (!(body->valid & OBD_MD_FLEASIZE))
1054 if (body->eadatasize == 0) {
1055 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1056 GOTO(out, rc = -EPROTO);
1059 /* The MDS sent back the EA because we unlinked the last reference
1060 * to this file. Use this EA to unlink the objects on the OST.
1061 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1062 * check it is complete and sensible. */
1063 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1065 LASSERT(eadata != NULL);
1067 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1069 CERROR("obd_unpackmd: %d\n", rc);
1072 LASSERT(rc >= sizeof(*lsm));
1076 GOTO(out_free_memmd, rc = -ENOMEM);
1078 oa->o_id = lsm->lsm_object_id;
1079 oa->o_seq = lsm->lsm_object_seq;
1080 oa->o_mode = body->mode & S_IFMT;
1081 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1083 if (body->valid & OBD_MD_FLCOOKIE) {
1084 oa->o_valid |= OBD_MD_FLCOOKIE;
1085 oti.oti_logcookies =
1086 req_capsule_server_sized_get(&request->rq_pill,
1088 sizeof(struct llog_cookie) *
1089 lsm->lsm_stripe_count);
1090 if (oti.oti_logcookies == NULL) {
1091 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1092 body->valid &= ~OBD_MD_FLCOOKIE;
1096 if (body->valid & OBD_MD_FLOSSCAPA) {
1097 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1099 GOTO(out_free_memmd, rc);
1102 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1106 CERROR("obd destroy objid "LPX64" error %d\n",
1107 lsm->lsm_object_id, rc);
1109 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1114 /* ll_unlink_generic() doesn't update the inode with the new link count.
1115 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1116 * is any lock existing. They will recycle dentries and inodes based upon locks
1118 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1119 struct dentry *dchild, struct qstr *name)
1121 struct ptlrpc_request *request = NULL;
1122 struct md_op_data *op_data;
1125 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1126 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1129 * XXX: unlink bind mountpoint maybe call to here,
1130 * just check it as vfs_unlink does.
1132 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1135 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1136 name->len, 0, LUSTRE_OPC_ANY, NULL);
1137 if (IS_ERR(op_data))
1138 RETURN(PTR_ERR(op_data));
1140 ll_get_child_fid(dir, name, &op_data->op_fid3);
1141 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1142 ll_finish_md_op_data(op_data);
1146 ll_update_times(request, dir);
1148 rc = ll_objects_destroy(request, dir);
1150 ptlrpc_req_finished(request);
1154 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1155 struct dentry *src_dchild, struct qstr *src_name,
1156 struct inode *tgt, struct dentry *tgt_dparent,
1157 struct dentry *tgt_dchild, struct qstr *tgt_name)
1159 struct ptlrpc_request *request = NULL;
1160 struct ll_sb_info *sbi = ll_i2sbi(src);
1161 struct md_op_data *op_data;
1164 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1165 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1166 src->i_ino, src->i_generation, src, tgt_name->len,
1167 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1169 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1170 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1173 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1174 LUSTRE_OPC_ANY, NULL);
1175 if (IS_ERR(op_data))
1176 RETURN(PTR_ERR(op_data));
1178 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1179 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1180 err = md_rename(sbi->ll_md_exp, op_data,
1181 src_name->name, src_name->len,
1182 tgt_name->name, tgt_name->len, &request);
1183 ll_finish_md_op_data(op_data);
1185 ll_update_times(request, src);
1186 ll_update_times(request, tgt);
1187 err = ll_objects_destroy(request, src);
1190 ptlrpc_req_finished(request);
1195 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1198 return ll_mknod_generic(dir, &dchild->d_name, mode,
1199 old_encode_dev(rdev), dchild);
1202 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1204 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1206 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1208 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1210 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1212 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1214 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1215 const char *oldname)
1217 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1219 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1220 struct dentry *new_dentry)
1222 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1225 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1226 struct inode *new_dir, struct dentry *new_dentry)
1229 err = ll_rename_generic(old_dir, NULL,
1230 old_dentry, &old_dentry->d_name,
1231 new_dir, NULL, new_dentry,
1232 &new_dentry->d_name);
1234 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1235 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1237 d_move(old_dentry, new_dentry);
1242 struct inode_operations ll_dir_inode_operations = {
1244 .lookup = ll_lookup_nd,
1245 .create = ll_create_nd,
1246 /* We need all these non-raw things for NFSD, to not patch it. */
1247 .unlink = ll_unlink,
1250 .symlink = ll_symlink,
1252 .rename = ll_rename,
1253 .setattr = ll_setattr,
1254 .getattr = ll_getattr,
1255 .permission = ll_inode_permission,
1256 .setxattr = ll_setxattr,
1257 .getxattr = ll_getxattr,
1258 .listxattr = ll_listxattr,
1259 .removexattr = ll_removexattr,
1262 struct inode_operations ll_special_inode_operations = {
1263 .setattr = ll_setattr,
1264 .getattr = ll_getattr,
1265 .permission = ll_inode_permission,
1266 .setxattr = ll_setxattr,
1267 .getxattr = ll_getxattr,
1268 .listxattr = ll_listxattr,
1269 .removexattr = ll_removexattr,