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, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
40 #include <linux/sched.h>
42 #include <linux/smp_lock.h>
43 #include <linux/quotaops.h>
44 #include <linux/highmem.h>
45 #include <linux/pagemap.h>
46 #include <linux/security.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;
116 if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
117 CERROR("Can not initialize inode "DFID" without object type: "
118 "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
122 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
123 if (unlikely(inode->i_mode == 0)) {
124 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
135 * Get an inode by inode number (already instantiated by the intent lookup).
136 * Returns inode or NULL
138 struct inode *ll_iget(struct super_block *sb, ino_t hash,
139 struct lustre_md *md)
145 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
148 if (inode->i_state & I_NEW) {
151 ll_read_inode2(inode, md);
152 if (S_ISREG(inode->i_mode) &&
153 ll_i2info(inode)->lli_clob == NULL)
154 rc = cl_file_inode_init(inode, md);
157 make_bad_inode(inode);
158 unlock_new_inode(inode);
162 unlock_new_inode(inode);
163 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
164 ll_update_inode(inode, md);
165 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
166 inode, PFID(&md->body->fid1));
171 static void ll_drop_negative_dentry(struct inode *dir)
173 struct dentry *dentry, *tmp_alias, *tmp_subdir;
175 cfs_spin_lock(&ll_lookup_lock);
176 spin_lock(&dcache_lock);
178 list_for_each_entry_safe(dentry, tmp_alias,
179 &dir->i_dentry,d_alias) {
180 if (!list_empty(&dentry->d_subdirs)) {
181 struct dentry *child;
182 list_for_each_entry_safe(child, tmp_subdir,
185 /* XXX Print some debug here? */
187 /* Negative dentry. If we were
188 dropping dcache lock, go
189 throught the list again */
190 if (ll_drop_dentry(child))
195 spin_unlock(&dcache_lock);
196 cfs_spin_unlock(&ll_lookup_lock);
200 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
201 void *data, int flag)
204 struct lustre_handle lockh;
208 case LDLM_CB_BLOCKING:
209 ldlm_lock2handle(lock, &lockh);
210 rc = ldlm_cli_cancel(&lockh);
212 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
216 case LDLM_CB_CANCELING: {
217 struct inode *inode = ll_inode_from_lock(lock);
218 struct ll_inode_info *lli;
219 __u64 bits = lock->l_policy_data.l_inodebits.bits;
221 ldlm_mode_t mode = lock->l_req_mode;
223 /* Invalidate all dentries associated with this inode */
227 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
228 /* For OPEN locks we differentiate between lock modes - CR, CW. PR - bug 22891 */
229 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE))
230 ll_have_md_lock(inode, &bits, LCK_MINMODE);
232 if (bits & MDS_INODELOCK_OPEN)
233 ll_have_md_lock(inode, &bits, mode);
235 fid = ll_inode2fid(inode);
236 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
237 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
238 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
239 LDLM_ERROR(lock, "data mismatch with object "
240 DFID" (%p)", PFID(fid), inode);
243 if (bits & MDS_INODELOCK_OPEN) {
245 switch (lock->l_req_mode) {
256 CERROR("Unexpected lock mode for OPEN lock "
257 "%d, inode %ld\n", lock->l_req_mode,
260 ll_md_real_close(inode, flags);
263 lli = ll_i2info(inode);
264 if (bits & MDS_INODELOCK_UPDATE)
265 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
267 if (S_ISDIR(inode->i_mode) &&
268 (bits & MDS_INODELOCK_UPDATE)) {
269 CDEBUG(D_INODE, "invalidating inode %lu\n",
271 truncate_inode_pages(inode->i_mapping, 0);
272 ll_drop_negative_dentry(inode);
275 if (inode->i_sb->s_root &&
276 inode != inode->i_sb->s_root->d_inode &&
277 (bits & MDS_INODELOCK_LOOKUP))
278 ll_unhash_aliases(inode);
289 __u32 ll_i2suppgid(struct inode *i)
291 if (cfs_curproc_is_in_groups(i->i_gid))
292 return (__u32)i->i_gid;
297 /* Pack the required supplementary groups into the supplied groups array.
298 * If we don't need to use the groups from the target inode(s) then we
299 * instead pack one or more groups from the user's supplementary group
300 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
301 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
308 LASSERT(suppgids != NULL);
310 suppgids[0] = ll_i2suppgid(i1);
313 suppgids[1] = ll_i2suppgid(i2);
318 for (i = 0; i < current_ngroups; i++) {
319 if (suppgids[0] == -1) {
320 if (current_groups[i] != suppgids[1])
321 suppgids[0] = current_groups[i];
324 if (suppgids[1] == -1) {
325 if (current_groups[i] != suppgids[0])
326 suppgids[1] = current_groups[i];
334 static void ll_d_add(struct dentry *de, struct inode *inode)
336 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
338 if (!list_empty(&de->d_alias)) {
339 spin_unlock(&dcache_lock);
340 CERROR("dentry %.*s %p alias next %p, prev %p\n",
341 de->d_name.len, de->d_name.name, de,
342 de->d_alias.next, de->d_alias.prev);
346 list_add(&de->d_alias, &inode->i_dentry);
348 /* d_instantiate() replacement code should initialize security
350 security_d_instantiate(de, inode);
353 if (!d_unhashed(de)) {
354 spin_unlock(&dcache_lock);
355 CERROR("dentry %.*s %p hash next %p\n",
356 de->d_name.len, de->d_name.name, de, de->d_hash.next);
359 d_rehash_cond(de, 0);
362 /* Search "inode"'s alias list for a dentry that has the same name and parent
363 * as de. If found, return it. If not found, return de.
364 * Lustre can't use d_add_unique because don't unhash aliases for directory
365 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
366 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
368 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
370 struct list_head *tmp;
371 struct dentry *dentry;
372 struct dentry *last_discon = NULL;
374 cfs_spin_lock(&ll_lookup_lock);
375 spin_lock(&dcache_lock);
376 list_for_each(tmp, &inode->i_dentry) {
377 dentry = list_entry(tmp, struct dentry, d_alias);
379 /* We are called here with 'de' already on the aliases list. */
380 if (unlikely(dentry == de)) {
385 if (dentry->d_flags & DCACHE_DISCONNECTED) {
386 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
387 last_discon = dentry;
391 if (dentry->d_parent != de->d_parent)
394 if (dentry->d_name.hash != de->d_name.hash)
397 if (dentry->d_name.len != de->d_name.len)
400 if (memcmp(dentry->d_name.name, de->d_name.name,
401 de->d_name.len) != 0)
405 ll_dops_init(dentry, 0, 1);
406 ll_dentry_rehash(dentry, 1);
407 spin_unlock(&dcache_lock);
408 cfs_spin_unlock(&ll_lookup_lock);
410 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
411 "refc %d\n", de->d_name.len, de->d_name.name, de,
412 de->d_parent, de->d_inode, atomic_read(&de->d_count));
417 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
418 "refc %d\n", last_discon, last_discon->d_inode,
419 atomic_read(&last_discon->d_count));
420 dget_locked(last_discon);
421 lock_dentry(last_discon);
422 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
423 unlock_dentry(last_discon);
424 spin_unlock(&dcache_lock);
425 cfs_spin_unlock(&ll_lookup_lock);
426 ll_dops_init(last_discon, 1, 1);
428 d_move(last_discon, de);
433 de->d_flags |= DCACHE_LUSTRE_INVALID;
437 spin_unlock(&dcache_lock);
438 cfs_spin_unlock(&ll_lookup_lock);
443 int ll_lookup_it_finish(struct ptlrpc_request *request,
444 struct lookup_intent *it, void *data)
446 struct it_cb_data *icbd = data;
447 struct dentry **de = icbd->icbd_childp;
448 struct inode *parent = icbd->icbd_parent;
449 struct inode *inode = NULL;
453 /* NB 1 request reference will be taken away by ll_intent_lock()
455 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
456 struct dentry *save = *de;
459 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
463 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
465 /* We used to query real size from OSTs here, but actually
466 this is not needed. For stat() calls size would be updated
467 from subsequent do_revalidate()->ll_inode_revalidate_it() in
469 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
470 Everybody else who needs correct file size would call
471 ll_glimpse_size or some equivalent themselves anyway.
472 Also see bug 7198. */
473 ll_dops_init(*de, 1, 1);
474 *de = ll_find_alias(inode, *de);
476 struct ll_dentry_data *lld = ll_d2d(*de);
478 /* just make sure the ll_dentry_data is ready */
479 if (unlikely(lld == NULL))
480 ll_dops_init(*de, 1, 1);
482 /* we have lookup look - unhide dentry */
483 ll_dentry_reset_flags(*de, bits);
487 ll_dops_init(*de, 1, 1);
488 /* Check that parent has UPDATE lock. If there is none, we
489 cannot afford to hash this dentry (done by ll_d_add) as it
490 might get picked up later when UPDATE lock will appear */
491 ibits = MDS_INODELOCK_UPDATE;
492 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE)) {
493 spin_lock(&dcache_lock);
495 spin_unlock(&dcache_lock);
497 /* negative lookup - and don't have update lock to
500 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
503 (*de)->d_inode = NULL;
504 /* We do not want to hash the dentry if don`t have a
505 * lock, but if this dentry is later used in d_move,
506 * we'd hit uninitialised list head d_hash, so we just
507 * do this to init d_hash field but leave dentry
508 * unhashed. (bug 10796). */
517 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
518 struct lookup_intent *it, int lookup_flags)
520 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
521 struct dentry *save = dentry, *retval;
522 struct ptlrpc_request *req = NULL;
523 struct md_op_data *op_data;
524 struct it_cb_data icbd;
529 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
530 RETURN(ERR_PTR(-ENAMETOOLONG));
532 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
533 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
534 parent->i_generation, parent, LL_IT2STR(it));
536 if (d_mountpoint(dentry))
537 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
539 ll_frob_intent(&it, &lookup_it);
541 /* As do_lookup is called before follow_mount, root dentry may be left
542 * not valid, revalidate it here. */
543 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
544 (it->it_op & (IT_OPEN | IT_CREAT))) {
545 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
546 MDS_INODELOCK_LOOKUP);
551 if (it->it_op == IT_GETATTR) {
552 rc = ll_statahead_enter(parent, &dentry, 0);
555 GOTO(out, retval = NULL);
556 GOTO(out, retval = dentry);
560 icbd.icbd_childp = &dentry;
561 icbd.icbd_parent = parent;
563 if (it->it_op & IT_CREAT ||
564 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
565 opc = LUSTRE_OPC_CREATE;
567 opc = LUSTRE_OPC_ANY;
569 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
570 dentry->d_name.len, lookup_flags, opc,
573 RETURN((void *)op_data);
575 it->it_create_mode &= ~cfs_curproc_umask();
577 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
578 lookup_flags, &req, ll_md_blocking_ast, 0);
579 ll_finish_md_op_data(op_data);
581 GOTO(out, retval = ERR_PTR(rc));
583 rc = ll_lookup_it_finish(req, it, &icbd);
585 ll_intent_release(it);
586 GOTO(out, retval = ERR_PTR(rc));
589 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
590 !S_ISREG(dentry->d_inode->i_mode) &&
591 !S_ISDIR(dentry->d_inode->i_mode)) {
592 ll_release_openhandle(dentry, it);
594 ll_lookup_finish_locks(it, dentry);
597 GOTO(out, retval = NULL);
599 GOTO(out, retval = dentry);
602 ptlrpc_req_finished(req);
603 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
604 ll_statahead_mark(parent, dentry);
608 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
611 struct lookup_intent *it;
613 OBD_ALLOC(it, sizeof(*it));
615 return ERR_PTR(-ENOMEM);
617 if (lookup_flags & LOOKUP_OPEN) {
619 if (lookup_flags & LOOKUP_CREATE)
620 it->it_op |= IT_CREAT;
621 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
622 it->it_flags = oit->flags;
624 it->it_op = IT_GETATTR;
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 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 if (S_ISFIFO(dentry->d_inode->i_mode)) {
668 // We cannot call open here as it would
671 (struct ptlrpc_request *)
672 it->d.lustre.it_data);
674 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
675 /* 2.6.1[456] have a bug in open_namei() that forgets to check
676 * nd->intent.open.file for error, so we need to return it as lookup's result
679 nd->intent.open.file->private_data = it;
680 filp =lookup_instantiate_filp(nd,dentry,
685 de = (struct dentry *) filp;
688 nd->intent.open.file->private_data = it;
689 (void)lookup_instantiate_filp(nd,dentry,
694 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
695 // XXX This can only reliably work on assumption
696 // that there are NO hashed negative dentries.
697 ll_d2d(dentry)->lld_it = it;
698 it = NULL; /* Will be freed in ll_create_nd */
699 /* We absolutely depend on ll_create_nd to be
700 * called to not leak this intent and possible
701 * data attached to it */
706 ll_intent_release(it);
707 OBD_FREE(it, sizeof(*it));
710 de = ll_lookup_it(parent, dentry, NULL, 0);
716 /* We depend on "mode" being set with the proper file type/umask by now */
717 static struct inode *ll_create_node(struct inode *dir, const char *name,
718 int namelen, const void *data, int datalen,
719 int mode, __u64 extra,
720 struct lookup_intent *it)
722 struct inode *inode = NULL;
723 struct ptlrpc_request *request = NULL;
724 struct ll_sb_info *sbi = ll_i2sbi(dir);
728 LASSERT(it && it->d.lustre.it_disposition);
730 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
731 request = it->d.lustre.it_data;
732 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
733 rc = ll_prep_inode(&inode, request, dir->i_sb);
735 GOTO(out, inode = ERR_PTR(rc));
737 LASSERT(list_empty(&inode->i_dentry));
739 /* We asked for a lock on the directory, but were granted a
740 * lock on the inode. Since we finally have an inode pointer,
741 * stuff it in the lock. */
742 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
743 inode, inode->i_ino, inode->i_generation);
744 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
747 ptlrpc_req_finished(request);
752 * By the time this is called, we already have created the directory cache
753 * entry for the new file, but it is so far negative - it has no inode.
755 * We defer creating the OBD object(s) until open, to keep the intent and
756 * non-intent code paths similar, and also because we do not have the MDS
757 * inode number before calling ll_create_node() (which is needed for LOV),
758 * so we would need to do yet another RPC to the MDS to store the LOV EA
759 * data on the MDS. If needed, we would pass the PACKED lmm as data and
760 * lmm_size in datalen (the MDS still has code which will handle that).
762 * If the create succeeds, we fill in the inode information
763 * with d_instantiate().
765 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
766 struct lookup_intent *it)
772 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
773 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
774 dir->i_generation, dir, LL_IT2STR(it));
776 rc = it_open_error(DISP_OPEN_CREATE, it);
780 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
781 NULL, 0, mode, 0, it);
783 RETURN(PTR_ERR(inode));
785 d_instantiate(dentry, inode);
786 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
787 * but some callers, e.g. do_coredump, expect dentry to be hashed after
788 * successful create. Hash it here. */
789 spin_lock(&dcache_lock);
790 if (d_unhashed(dentry))
791 d_rehash_cond(dentry, 0);
792 spin_unlock(&dcache_lock);
796 static void ll_update_times(struct ptlrpc_request *request,
799 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
803 if (body->valid & OBD_MD_FLMTIME &&
804 body->mtime > LTIME_S(inode->i_mtime)) {
805 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
806 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
807 LTIME_S(inode->i_mtime) = body->mtime;
809 if (body->valid & OBD_MD_FLCTIME &&
810 body->ctime > LTIME_S(inode->i_ctime))
811 LTIME_S(inode->i_ctime) = body->ctime;
814 static int ll_new_node(struct inode *dir, struct qstr *name,
815 const char *tgt, int mode, int rdev,
816 struct dentry *dchild, __u32 opc)
818 struct ptlrpc_request *request = NULL;
819 struct md_op_data *op_data;
820 struct inode *inode = NULL;
821 struct ll_sb_info *sbi = ll_i2sbi(dir);
826 if (unlikely(tgt != NULL))
827 tgt_len = strlen(tgt) + 1;
829 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
830 name->len, 0, opc, NULL);
832 GOTO(err_exit, err = PTR_ERR(op_data));
834 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
835 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
836 cfs_curproc_cap_pack(), rdev, &request);
837 ll_finish_md_op_data(op_data);
841 ll_update_times(request, dir);
844 err = ll_prep_inode(&inode, request, dchild->d_sb);
849 d_instantiate(dchild, inode);
853 ptlrpc_req_finished(request);
858 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
859 unsigned rdev, struct dentry *dchild)
864 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
865 name->len, name->name, dir->i_ino, dir->i_generation, dir,
868 mode &= ~cfs_curproc_umask();
870 switch (mode & S_IFMT) {
872 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
878 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
889 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
894 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
895 int mode, struct nameidata *nd)
897 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
901 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
903 ll_d2d(dentry)->lld_it = NULL;
905 /* Was there an error? Propagate it! */
906 if (it->d.lustre.it_status) {
907 rc = it->d.lustre.it_status;
911 rc = ll_create_it(dir, dentry, mode, it);
912 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
913 nd->intent.open.file->private_data = it;
914 lookup_instantiate_filp(nd, dentry, NULL);
918 ll_intent_release(it);
919 OBD_FREE(it, sizeof(*it));
922 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
927 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
928 const char *tgt, struct dentry *dchild)
933 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
934 name->len, name->name, dir->i_ino, dir->i_generation,
937 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
938 0, dchild, LUSTRE_OPC_SYMLINK);
941 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
946 static int ll_link_generic(struct inode *src, struct inode *dir,
947 struct qstr *name, struct dentry *dchild)
949 struct ll_sb_info *sbi = ll_i2sbi(dir);
950 struct ptlrpc_request *request = NULL;
951 struct md_op_data *op_data;
956 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
957 src->i_ino, src->i_generation, src, dir->i_ino,
958 dir->i_generation, dir, name->len, name->name);
960 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
961 0, LUSTRE_OPC_ANY, NULL);
963 RETURN(PTR_ERR(op_data));
965 err = md_link(sbi->ll_md_exp, op_data, &request);
966 ll_finish_md_op_data(op_data);
972 ll_update_times(request, dir);
973 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
976 ptlrpc_req_finished(request);
980 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
981 int mode, struct dentry *dchild)
987 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
988 name->len, name->name, dir->i_ino, dir->i_generation, dir);
990 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
991 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
994 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
999 /* Try to find the child dentry by its name.
1000 If found, put the result fid into @fid. */
1001 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1004 struct dentry *parent, *child;
1006 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1007 child = d_lookup(parent, name);
1010 *fid = *ll_inode2fid(child->d_inode);
1015 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1016 struct dentry *dchild, struct qstr *name)
1018 struct ptlrpc_request *request = NULL;
1019 struct md_op_data *op_data;
1023 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1024 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1026 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1029 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1030 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1031 if (IS_ERR(op_data))
1032 RETURN(PTR_ERR(op_data));
1034 ll_get_child_fid(dir, name, &op_data->op_fid3);
1035 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1036 ll_finish_md_op_data(op_data);
1038 ll_update_times(request, dir);
1039 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1042 ptlrpc_req_finished(request);
1046 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1048 struct mdt_body *body;
1049 struct lov_mds_md *eadata;
1050 struct lov_stripe_md *lsm = NULL;
1051 struct obd_trans_info oti = { 0 };
1053 struct obd_capa *oc = NULL;
1057 /* req is swabbed so this is safe */
1058 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1059 if (!(body->valid & OBD_MD_FLEASIZE))
1062 if (body->eadatasize == 0) {
1063 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1064 GOTO(out, rc = -EPROTO);
1067 /* The MDS sent back the EA because we unlinked the last reference
1068 * to this file. Use this EA to unlink the objects on the OST.
1069 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1070 * check it is complete and sensible. */
1071 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1073 LASSERT(eadata != NULL);
1075 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1077 CERROR("obd_unpackmd: %d\n", rc);
1080 LASSERT(rc >= sizeof(*lsm));
1084 GOTO(out_free_memmd, rc = -ENOMEM);
1086 oa->o_id = lsm->lsm_object_id;
1087 oa->o_seq = lsm->lsm_object_seq;
1088 oa->o_mode = body->mode & S_IFMT;
1089 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1091 if (body->valid & OBD_MD_FLCOOKIE) {
1092 oa->o_valid |= OBD_MD_FLCOOKIE;
1093 oti.oti_logcookies =
1094 req_capsule_server_sized_get(&request->rq_pill,
1096 sizeof(struct llog_cookie) *
1097 lsm->lsm_stripe_count);
1098 if (oti.oti_logcookies == NULL) {
1099 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1100 body->valid &= ~OBD_MD_FLCOOKIE;
1104 if (body->valid & OBD_MD_FLOSSCAPA) {
1105 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1107 GOTO(out_free_memmd, rc);
1110 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, 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,