4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 #include <linux/sched.h>
40 #include <linux/smp_lock.h>
41 #include <linux/quotaops.h>
42 #include <linux/highmem.h>
43 #include <linux/pagemap.h>
44 #include <linux/security.h>
46 #define DEBUG_SUBSYSTEM S_LLITE
48 #include <obd_support.h>
49 #include <lustre_fid.h>
50 #include <lustre_lite.h>
51 #include <lustre_dlm.h>
52 #include <lustre_ver.h>
53 #include "llite_internal.h"
56 * Check if we have something mounted at the named dchild.
57 * In such a case there would always be dentry present.
59 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
64 if (unlikely(dchild)) {
65 mounted = d_mountpoint(dchild);
67 dchild = d_lookup(dparent, name);
69 mounted = d_mountpoint(dchild);
76 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
80 ldlm_lock_decref(lockh, mode);
86 /* called from iget5_locked->find_inode() under inode_lock spinlock */
87 static int ll_test_inode(struct inode *inode, void *opaque)
89 struct ll_inode_info *lli = ll_i2info(inode);
90 struct lustre_md *md = opaque;
92 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
93 CERROR("MDS body missing FID\n");
97 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
103 static int ll_set_inode(struct inode *inode, void *opaque)
105 struct ll_inode_info *lli = ll_i2info(inode);
106 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
108 if (unlikely(!(body->valid & OBD_MD_FLID))) {
109 CERROR("MDS body missing FID\n");
113 lli->lli_fid = body->fid1;
114 if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
115 CERROR("Can not initialize inode "DFID" without object type: "
116 "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
120 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
121 if (unlikely(inode->i_mode == 0)) {
122 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
133 * Get an inode by inode number (already instantiated by the intent lookup).
134 * Returns inode or NULL
136 struct inode *ll_iget(struct super_block *sb, ino_t hash,
137 struct lustre_md *md)
143 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
146 if (inode->i_state & I_NEW) {
149 ll_read_inode2(inode, md);
150 if (S_ISREG(inode->i_mode) &&
151 ll_i2info(inode)->lli_clob == NULL)
152 rc = cl_file_inode_init(inode, md);
155 make_bad_inode(inode);
156 unlock_new_inode(inode);
160 unlock_new_inode(inode);
161 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
162 ll_update_inode(inode, md);
163 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
164 inode, PFID(&md->body->fid1));
169 static void ll_drop_negative_dentry(struct inode *dir)
171 struct dentry *dentry, *tmp_alias, *tmp_subdir;
173 cfs_spin_lock(&ll_lookup_lock);
174 spin_lock(&dcache_lock);
176 list_for_each_entry_safe(dentry, tmp_alias,
177 &dir->i_dentry,d_alias) {
178 if (!list_empty(&dentry->d_subdirs)) {
179 struct dentry *child;
180 list_for_each_entry_safe(child, tmp_subdir,
183 /* XXX Print some debug here? */
185 /* Negative dentry. If we were
186 dropping dcache lock, go
187 throught the list again */
188 if (ll_drop_dentry(child))
193 spin_unlock(&dcache_lock);
194 cfs_spin_unlock(&ll_lookup_lock);
198 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
199 void *data, int flag)
202 struct lustre_handle lockh;
206 case LDLM_CB_BLOCKING:
207 ldlm_lock2handle(lock, &lockh);
208 rc = ldlm_cli_cancel(&lockh);
210 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
214 case LDLM_CB_CANCELING: {
215 struct inode *inode = ll_inode_from_lock(lock);
216 struct ll_inode_info *lli;
217 __u64 bits = lock->l_policy_data.l_inodebits.bits;
219 ldlm_mode_t mode = lock->l_req_mode;
221 /* Invalidate all dentries associated with this inode */
225 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
226 /* For OPEN locks we differentiate between lock modes - CR, CW. PR - bug 22891 */
227 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE))
228 ll_have_md_lock(inode, &bits, LCK_MINMODE);
230 if (bits & MDS_INODELOCK_OPEN)
231 ll_have_md_lock(inode, &bits, mode);
233 fid = ll_inode2fid(inode);
234 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
235 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
236 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
237 LDLM_ERROR(lock, "data mismatch with object "
238 DFID" (%p)", PFID(fid), inode);
241 if (bits & MDS_INODELOCK_OPEN) {
243 switch (lock->l_req_mode) {
254 CERROR("Unexpected lock mode for OPEN lock "
255 "%d, inode %ld\n", lock->l_req_mode,
258 ll_md_real_close(inode, flags);
261 lli = ll_i2info(inode);
262 if (bits & MDS_INODELOCK_UPDATE)
263 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
265 if (S_ISDIR(inode->i_mode) &&
266 (bits & MDS_INODELOCK_UPDATE)) {
267 CDEBUG(D_INODE, "invalidating inode %lu\n",
269 truncate_inode_pages(inode->i_mapping, 0);
270 ll_drop_negative_dentry(inode);
273 if (inode->i_sb->s_root &&
274 inode != inode->i_sb->s_root->d_inode &&
275 (bits & MDS_INODELOCK_LOOKUP))
276 ll_unhash_aliases(inode);
287 __u32 ll_i2suppgid(struct inode *i)
289 if (cfs_curproc_is_in_groups(i->i_gid))
290 return (__u32)i->i_gid;
295 /* Pack the required supplementary groups into the supplied groups array.
296 * If we don't need to use the groups from the target inode(s) then we
297 * instead pack one or more groups from the user's supplementary group
298 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
299 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
306 LASSERT(suppgids != NULL);
308 suppgids[0] = ll_i2suppgid(i1);
311 suppgids[1] = ll_i2suppgid(i2);
316 for (i = 0; i < current_ngroups; i++) {
317 if (suppgids[0] == -1) {
318 if (current_groups[i] != suppgids[1])
319 suppgids[0] = current_groups[i];
322 if (suppgids[1] == -1) {
323 if (current_groups[i] != suppgids[0])
324 suppgids[1] = current_groups[i];
332 static void ll_d_add(struct dentry *de, struct inode *inode)
334 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
336 if (!list_empty(&de->d_alias)) {
337 spin_unlock(&dcache_lock);
338 CERROR("dentry %.*s %p alias next %p, prev %p\n",
339 de->d_name.len, de->d_name.name, de,
340 de->d_alias.next, de->d_alias.prev);
344 list_add(&de->d_alias, &inode->i_dentry);
348 if (!d_unhashed(de)) {
349 spin_unlock(&dcache_lock);
350 CERROR("dentry %.*s %p hash next %p\n",
351 de->d_name.len, de->d_name.name, de, de->d_hash.next);
356 /* Search "inode"'s alias list for a dentry that has the same name and parent
357 * as de. If found, return it. If not found, return de.
358 * Lustre can't use d_add_unique because don't unhash aliases for directory
359 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
360 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
362 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
364 struct list_head *tmp;
365 struct dentry *dentry;
366 struct dentry *last_discon = NULL;
368 cfs_spin_lock(&ll_lookup_lock);
369 spin_lock(&dcache_lock);
370 list_for_each(tmp, &inode->i_dentry) {
371 dentry = list_entry(tmp, struct dentry, d_alias);
373 /* We are called here with 'de' already on the aliases list. */
374 if (unlikely(dentry == de)) {
379 if (dentry->d_flags & DCACHE_DISCONNECTED) {
380 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
381 last_discon = dentry;
385 if (dentry->d_parent != de->d_parent)
388 if (dentry->d_name.hash != de->d_name.hash)
391 if (dentry->d_name.len != de->d_name.len)
394 if (memcmp(dentry->d_name.name, de->d_name.name,
395 de->d_name.len) != 0)
399 ll_dops_init(dentry, 0, 1);
400 ll_dentry_rehash(dentry, 1);
401 spin_unlock(&dcache_lock);
402 cfs_spin_unlock(&ll_lookup_lock);
404 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
405 "refc %d\n", de->d_name.len, de->d_name.name, de,
406 de->d_parent, de->d_inode, atomic_read(&de->d_count));
411 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
412 "refc %d\n", last_discon, last_discon->d_inode,
413 atomic_read(&last_discon->d_count));
414 dget_locked(last_discon);
415 lock_dentry(last_discon);
416 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
417 unlock_dentry(last_discon);
418 spin_unlock(&dcache_lock);
419 cfs_spin_unlock(&ll_lookup_lock);
420 ll_dops_init(last_discon, 1, 1);
422 d_move(last_discon, de);
427 de->d_flags |= DCACHE_LUSTRE_INVALID;
430 spin_unlock(&dcache_lock);
431 cfs_spin_unlock(&ll_lookup_lock);
433 security_d_instantiate(de, inode);
439 int ll_lookup_it_finish(struct ptlrpc_request *request,
440 struct lookup_intent *it, void *data)
442 struct it_cb_data *icbd = data;
443 struct dentry **de = icbd->icbd_childp;
444 struct inode *parent = icbd->icbd_parent;
445 struct inode *inode = NULL;
449 /* NB 1 request reference will be taken away by ll_intent_lock()
451 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
452 struct dentry *save = *de;
455 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
459 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
461 /* We used to query real size from OSTs here, but actually
462 this is not needed. For stat() calls size would be updated
463 from subsequent do_revalidate()->ll_inode_revalidate_it() in
465 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
466 Everybody else who needs correct file size would call
467 ll_glimpse_size or some equivalent themselves anyway.
468 Also see bug 7198. */
469 ll_dops_init(*de, 1, 1);
470 *de = ll_find_alias(inode, *de);
472 struct ll_dentry_data *lld = ll_d2d(*de);
474 /* just make sure the ll_dentry_data is ready */
475 if (unlikely(lld == NULL))
476 ll_dops_init(*de, 1, 1);
478 /* we have lookup look - unhide dentry */
479 ll_dentry_reset_flags(*de, bits);
481 struct lookup_intent parent_it = {
483 .d.lustre.it_lock_handle = 0 };
485 ll_dops_init(*de, 1, 1);
486 /* Check that parent has UPDATE lock. If there is none, we
487 cannot afford to hash this dentry (done by ll_d_add) as it
488 might get picked up later when UPDATE lock will appear;
489 otherwise, add ref to the parent UPDATE lock, to make sure
490 ll_d_add() undergoes with parent's UPDATE lock held */
491 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
492 &ll_i2info(parent)->lli_fid, NULL)) {
493 spin_lock(&dcache_lock);
495 spin_unlock(&dcache_lock);
496 security_d_instantiate(*de, inode);
498 ll_intent_release(&parent_it);
500 /* negative lookup - and don't have update lock to
503 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
506 (*de)->d_inode = NULL;
507 /* We do not want to hash the dentry if don`t have a
508 * lock, but if this dentry is later used in d_move,
509 * we'd hit uninitialised list head d_hash, so we just
510 * do this to init d_hash field but leave dentry
511 * unhashed. (bug 10796). */
520 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
521 struct lookup_intent *it, int lookup_flags)
523 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
524 struct dentry *save = dentry, *retval;
525 struct ptlrpc_request *req = NULL;
526 struct md_op_data *op_data;
527 struct it_cb_data icbd;
532 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
533 RETURN(ERR_PTR(-ENAMETOOLONG));
535 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
536 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
537 parent->i_generation, parent, LL_IT2STR(it));
539 if (d_mountpoint(dentry))
540 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
542 ll_frob_intent(&it, &lookup_it);
544 /* As do_lookup is called before follow_mount, root dentry may be left
545 * not valid, revalidate it here. */
546 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
547 (it->it_op & (IT_OPEN | IT_CREAT))) {
548 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
549 MDS_INODELOCK_LOOKUP);
554 if (it->it_op == IT_GETATTR) {
555 rc = ll_statahead_enter(parent, &dentry, 0);
558 GOTO(out, retval = NULL);
559 GOTO(out, retval = dentry);
563 icbd.icbd_childp = &dentry;
564 icbd.icbd_parent = parent;
566 if (it->it_op & IT_CREAT ||
567 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
568 opc = LUSTRE_OPC_CREATE;
570 opc = LUSTRE_OPC_ANY;
572 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
573 dentry->d_name.len, lookup_flags, opc,
576 RETURN((void *)op_data);
578 it->it_create_mode &= ~cfs_curproc_umask();
580 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
581 lookup_flags, &req, ll_md_blocking_ast, 0);
582 ll_finish_md_op_data(op_data);
584 GOTO(out, retval = ERR_PTR(rc));
586 rc = ll_lookup_it_finish(req, it, &icbd);
588 ll_intent_release(it);
589 GOTO(out, retval = ERR_PTR(rc));
592 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
593 !S_ISREG(dentry->d_inode->i_mode) &&
594 !S_ISDIR(dentry->d_inode->i_mode)) {
595 ll_release_openhandle(dentry, it);
597 ll_lookup_finish_locks(it, dentry);
600 GOTO(out, retval = NULL);
602 GOTO(out, retval = dentry);
605 ptlrpc_req_finished(req);
606 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
607 ll_statahead_mark(parent, dentry);
611 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
614 struct lookup_intent *it;
616 OBD_ALLOC(it, sizeof(*it));
618 return ERR_PTR(-ENOMEM);
620 if (lookup_flags & LOOKUP_OPEN) {
622 if (lookup_flags & LOOKUP_CREATE)
623 it->it_op |= IT_CREAT;
624 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
625 it->it_flags = oit->flags;
627 it->it_op = IT_GETATTR;
633 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
634 struct nameidata *nd)
639 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
640 struct lookup_intent *it;
642 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17)
643 /* Did we came here from failed revalidate just to propagate
645 if (nd->flags & LOOKUP_OPEN)
646 if (IS_ERR(nd->intent.open.file))
647 RETURN((struct dentry *)nd->intent.open.file);
649 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
650 it = ll_d2d(dentry)->lld_it;
651 ll_d2d(dentry)->lld_it = NULL;
653 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
654 /* We are sure this is new dentry, so we need to create
655 our private data and set the dentry ops */
656 ll_dops_init(dentry, 1, 1);
659 it = ll_convert_intent(&nd->intent.open, nd->flags);
661 RETURN((struct dentry *)it);
664 de = ll_lookup_it(parent, dentry, it, nd->flags);
667 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
668 if (dentry->d_inode &&
669 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
670 if (S_ISFIFO(dentry->d_inode->i_mode)) {
671 // We cannot call open here as it would
674 (struct ptlrpc_request *)
675 it->d.lustre.it_data);
679 nd->intent.open.file->private_data = it;
680 filp = lookup_instantiate_filp(nd,
686 de = (struct dentry *)filp;
689 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
690 // XXX This can only reliably work on assumption
691 // that there are NO hashed negative dentries.
692 ll_d2d(dentry)->lld_it = it;
693 it = NULL; /* Will be freed in ll_create_nd */
694 /* We absolutely depend on ll_create_nd to be
695 * called to not leak this intent and possible
696 * data attached to it */
701 ll_intent_release(it);
702 OBD_FREE(it, sizeof(*it));
705 de = ll_lookup_it(parent, dentry, NULL, 0);
711 /* We depend on "mode" being set with the proper file type/umask by now */
712 static struct inode *ll_create_node(struct inode *dir, const char *name,
713 int namelen, const void *data, int datalen,
714 int mode, __u64 extra,
715 struct lookup_intent *it)
717 struct inode *inode = NULL;
718 struct ptlrpc_request *request = NULL;
719 struct ll_sb_info *sbi = ll_i2sbi(dir);
723 LASSERT(it && it->d.lustre.it_disposition);
725 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
726 request = it->d.lustre.it_data;
727 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
728 rc = ll_prep_inode(&inode, request, dir->i_sb);
730 GOTO(out, inode = ERR_PTR(rc));
732 LASSERT(list_empty(&inode->i_dentry));
734 /* We asked for a lock on the directory, but were granted a
735 * lock on the inode. Since we finally have an inode pointer,
736 * stuff it in the lock. */
737 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
738 inode, inode->i_ino, inode->i_generation);
739 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
742 ptlrpc_req_finished(request);
747 * By the time this is called, we already have created the directory cache
748 * entry for the new file, but it is so far negative - it has no inode.
750 * We defer creating the OBD object(s) until open, to keep the intent and
751 * non-intent code paths similar, and also because we do not have the MDS
752 * inode number before calling ll_create_node() (which is needed for LOV),
753 * so we would need to do yet another RPC to the MDS to store the LOV EA
754 * data on the MDS. If needed, we would pass the PACKED lmm as data and
755 * lmm_size in datalen (the MDS still has code which will handle that).
757 * If the create succeeds, we fill in the inode information
758 * with d_instantiate().
760 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
761 struct lookup_intent *it)
767 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
768 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
769 dir->i_generation, dir, LL_IT2STR(it));
771 rc = it_open_error(DISP_OPEN_CREATE, it);
775 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
776 NULL, 0, mode, 0, it);
778 RETURN(PTR_ERR(inode));
780 d_instantiate(dentry, inode);
781 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
782 * but some callers, e.g. do_coredump, expect dentry to be hashed after
783 * successful create. Hash it here. */
784 spin_lock(&dcache_lock);
785 if (d_unhashed(dentry))
786 d_rehash_cond(dentry, 0);
787 spin_unlock(&dcache_lock);
791 static void ll_update_times(struct ptlrpc_request *request,
794 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
798 if (body->valid & OBD_MD_FLMTIME &&
799 body->mtime > LTIME_S(inode->i_mtime)) {
800 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
801 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
802 LTIME_S(inode->i_mtime) = body->mtime;
804 if (body->valid & OBD_MD_FLCTIME &&
805 body->ctime > LTIME_S(inode->i_ctime))
806 LTIME_S(inode->i_ctime) = body->ctime;
809 static int ll_new_node(struct inode *dir, struct qstr *name,
810 const char *tgt, int mode, int rdev,
811 struct dentry *dchild, __u32 opc)
813 struct ptlrpc_request *request = NULL;
814 struct md_op_data *op_data;
815 struct inode *inode = NULL;
816 struct ll_sb_info *sbi = ll_i2sbi(dir);
821 if (unlikely(tgt != NULL))
822 tgt_len = strlen(tgt) + 1;
824 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
825 name->len, 0, opc, NULL);
827 GOTO(err_exit, err = PTR_ERR(op_data));
829 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
830 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
831 cfs_curproc_cap_pack(), rdev, &request);
832 ll_finish_md_op_data(op_data);
836 ll_update_times(request, dir);
839 err = ll_prep_inode(&inode, request, dchild->d_sb);
844 d_instantiate(dchild, inode);
848 ptlrpc_req_finished(request);
853 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
854 unsigned rdev, struct dentry *dchild)
859 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
860 name->len, name->name, dir->i_ino, dir->i_generation, dir,
863 mode &= ~cfs_curproc_umask();
865 switch (mode & S_IFMT) {
867 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
873 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
884 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
889 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
890 int mode, struct nameidata *nd)
892 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
896 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
898 ll_d2d(dentry)->lld_it = NULL;
900 /* Was there an error? Propagate it! */
901 if (it->d.lustre.it_status) {
902 rc = it->d.lustre.it_status;
906 rc = ll_create_it(dir, dentry, mode, it);
907 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
910 nd->intent.open.file->private_data = it;
911 filp = lookup_instantiate_filp(nd, dentry, NULL);
917 ll_intent_release(it);
918 OBD_FREE(it, sizeof(*it));
921 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
926 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
927 const char *tgt, struct dentry *dchild)
932 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
933 name->len, name->name, dir->i_ino, dir->i_generation,
936 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
937 0, dchild, LUSTRE_OPC_SYMLINK);
940 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
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);
972 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
975 ptlrpc_req_finished(request);
979 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
980 int mode, struct dentry *dchild)
986 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
987 name->len, name->name, dir->i_ino, dir->i_generation, dir);
989 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
990 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
993 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
998 /* Try to find the child dentry by its name.
999 If found, put the result fid into @fid. */
1000 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1003 struct dentry *parent, *child;
1005 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1006 child = d_lookup(parent, name);
1009 *fid = *ll_inode2fid(child->d_inode);
1014 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1015 struct dentry *dchild, struct qstr *name)
1017 struct ptlrpc_request *request = NULL;
1018 struct md_op_data *op_data;
1022 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1023 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1025 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1028 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1029 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1030 if (IS_ERR(op_data))
1031 RETURN(PTR_ERR(op_data));
1033 ll_get_child_fid(dir, name, &op_data->op_fid3);
1034 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1035 ll_finish_md_op_data(op_data);
1037 ll_update_times(request, dir);
1038 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1041 ptlrpc_req_finished(request);
1045 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1047 struct mdt_body *body;
1048 struct lov_mds_md *eadata;
1049 struct lov_stripe_md *lsm = NULL;
1050 struct obd_trans_info oti = { 0 };
1052 struct obd_capa *oc = NULL;
1056 /* req is swabbed so this is safe */
1057 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1058 if (!(body->valid & OBD_MD_FLEASIZE))
1061 if (body->eadatasize == 0) {
1062 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1063 GOTO(out, rc = -EPROTO);
1066 /* The MDS sent back the EA because we unlinked the last reference
1067 * to this file. Use this EA to unlink the objects on the OST.
1068 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1069 * check it is complete and sensible. */
1070 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1072 LASSERT(eadata != NULL);
1074 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1076 CERROR("obd_unpackmd: %d\n", rc);
1079 LASSERT(rc >= sizeof(*lsm));
1083 GOTO(out_free_memmd, rc = -ENOMEM);
1085 oa->o_id = lsm->lsm_object_id;
1086 oa->o_seq = lsm->lsm_object_seq;
1087 oa->o_mode = body->mode & S_IFMT;
1088 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1090 if (body->valid & OBD_MD_FLCOOKIE) {
1091 oa->o_valid |= OBD_MD_FLCOOKIE;
1092 oti.oti_logcookies =
1093 req_capsule_server_sized_get(&request->rq_pill,
1095 sizeof(struct llog_cookie) *
1096 lsm->lsm_stripe_count);
1097 if (oti.oti_logcookies == NULL) {
1098 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1099 body->valid &= ~OBD_MD_FLCOOKIE;
1103 if (body->valid & OBD_MD_FLOSSCAPA) {
1104 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1106 GOTO(out_free_memmd, rc);
1109 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1110 ll_i2mdexp(dir), oc);
1114 CERROR("obd destroy objid "LPX64" error %d\n",
1115 lsm->lsm_object_id, rc);
1117 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1122 /* ll_unlink_generic() doesn't update the inode with the new link count.
1123 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1124 * is any lock existing. They will recycle dentries and inodes based upon locks
1126 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1127 struct dentry *dchild, struct qstr *name)
1129 struct ptlrpc_request *request = NULL;
1130 struct md_op_data *op_data;
1133 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1134 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1137 * XXX: unlink bind mountpoint maybe call to here,
1138 * just check it as vfs_unlink does.
1140 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1143 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1144 name->len, 0, LUSTRE_OPC_ANY, NULL);
1145 if (IS_ERR(op_data))
1146 RETURN(PTR_ERR(op_data));
1148 ll_get_child_fid(dir, name, &op_data->op_fid3);
1149 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1150 ll_finish_md_op_data(op_data);
1154 ll_update_times(request, dir);
1155 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1157 rc = ll_objects_destroy(request, dir);
1159 ptlrpc_req_finished(request);
1163 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1164 struct dentry *src_dchild, struct qstr *src_name,
1165 struct inode *tgt, struct dentry *tgt_dparent,
1166 struct dentry *tgt_dchild, struct qstr *tgt_name)
1168 struct ptlrpc_request *request = NULL;
1169 struct ll_sb_info *sbi = ll_i2sbi(src);
1170 struct md_op_data *op_data;
1173 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1174 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1175 src->i_ino, src->i_generation, src, tgt_name->len,
1176 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1178 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1179 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1182 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1183 LUSTRE_OPC_ANY, NULL);
1184 if (IS_ERR(op_data))
1185 RETURN(PTR_ERR(op_data));
1187 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1188 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1189 err = md_rename(sbi->ll_md_exp, op_data,
1190 src_name->name, src_name->len,
1191 tgt_name->name, tgt_name->len, &request);
1192 ll_finish_md_op_data(op_data);
1194 ll_update_times(request, src);
1195 ll_update_times(request, tgt);
1196 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1197 err = ll_objects_destroy(request, src);
1200 ptlrpc_req_finished(request);
1205 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1208 return ll_mknod_generic(dir, &dchild->d_name, mode,
1209 old_encode_dev(rdev), dchild);
1212 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1214 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1216 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1218 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1220 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1222 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1224 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1225 const char *oldname)
1227 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1229 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1230 struct dentry *new_dentry)
1232 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1235 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1236 struct inode *new_dir, struct dentry *new_dentry)
1239 err = ll_rename_generic(old_dir, NULL,
1240 old_dentry, &old_dentry->d_name,
1241 new_dir, NULL, new_dentry,
1242 &new_dentry->d_name);
1244 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1245 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1247 d_move(old_dentry, new_dentry);
1252 struct inode_operations ll_dir_inode_operations = {
1254 .lookup = ll_lookup_nd,
1255 .create = ll_create_nd,
1256 /* We need all these non-raw things for NFSD, to not patch it. */
1257 .unlink = ll_unlink,
1260 .symlink = ll_symlink,
1262 .rename = ll_rename,
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,
1272 struct inode_operations ll_special_inode_operations = {
1273 .setattr = ll_setattr,
1274 .getattr = ll_getattr,
1275 .permission = ll_inode_permission,
1276 .setxattr = ll_setxattr,
1277 .getxattr = ll_getxattr,
1278 .listxattr = ll_listxattr,
1279 .removexattr = ll_removexattr,