1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #include <linux/sched.h>
25 #include <linux/smp_lock.h>
26 #include <linux/quotaops.h>
27 #include <linux/highmem.h>
28 #include <linux/pagemap.h>
30 #define DEBUG_SUBSYSTEM S_LLITE
32 #include <obd_support.h>
33 #include <lustre_fid.h>
34 #include <lustre_lite.h>
35 #include <lustre_dlm.h>
36 #include <lustre_ver.h>
37 #include <lustre_mdc.h>
38 #include "llite_internal.h"
41 extern struct dentry_operations ll_d_ops;
44 * Check if we have something mounted at the named dchild.
45 * In such a case there would always be dentry present.
47 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
52 if (unlikely(dchild)) {
53 mounted = d_mountpoint(dchild);
55 dchild = d_lookup(dparent, name);
57 mounted = d_mountpoint(dchild);
64 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
68 ldlm_lock_decref(lockh, mode);
74 * Get an inode by inode number (already instantiated by the intent lookup).
75 * Returns inode or NULL
77 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
78 struct inode *ll_iget(struct super_block *sb, ino_t hash,
81 struct ll_inode_info *lli;
85 inode = iget_locked(sb, hash);
87 if (inode->i_state & I_NEW) {
88 lli = ll_i2info(inode);
89 ll_read_inode2(inode, md);
90 unlock_new_inode(inode);
92 if (!(inode->i_state & (I_FREEING | I_CLEAR)))
93 ll_update_inode(inode, md);
95 CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n",
96 inode->i_ino, inode->i_generation, inode);
102 struct inode *ll_iget(struct super_block *sb, ino_t hash,
103 struct lustre_md *md)
108 inode = iget4(sb, hash, NULL, md);
110 if (!(inode->i_state & (I_FREEING | I_CLEAR)))
111 ll_update_inode(inode, md);
113 CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n",
114 inode->i_ino, inode->i_generation, inode);
120 static void ll_drop_negative_dentry(struct inode *dir)
122 struct dentry *dentry, *tmp_alias, *tmp_subdir;
124 spin_lock(&dcache_lock);
126 list_for_each_entry_safe(dentry, tmp_alias,
127 &dir->i_dentry,d_alias) {
128 if (!list_empty(&dentry->d_subdirs)) {
129 struct dentry *child;
130 list_for_each_entry_safe(child, tmp_subdir,
133 /* XXX Print some debug here? */
135 /* Negative dentry. If we were
136 dropping dcache lock, go
137 throught the list again */
138 if (ll_drop_dentry(child))
143 spin_unlock(&dcache_lock);
147 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
148 void *data, int flag)
151 struct lustre_handle lockh;
155 case LDLM_CB_BLOCKING:
156 ldlm_lock2handle(lock, &lockh);
157 rc = ldlm_cli_cancel(&lockh);
159 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
163 case LDLM_CB_CANCELING: {
164 struct inode *inode = ll_inode_from_lock(lock);
165 __u64 bits = lock->l_policy_data.l_inodebits.bits;
168 /* Invalidate all dentries associated with this inode */
172 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
173 if ((bits & MDS_INODELOCK_LOOKUP) &&
174 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
175 bits &= ~MDS_INODELOCK_LOOKUP;
176 if ((bits & MDS_INODELOCK_UPDATE) &&
177 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
178 bits &= ~MDS_INODELOCK_UPDATE;
179 if ((bits & MDS_INODELOCK_OPEN) &&
180 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
181 bits &= ~MDS_INODELOCK_OPEN;
183 fid = ll_inode2fid(inode);
184 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
185 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
186 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
187 LDLM_ERROR(lock, "data mismatch with object "
188 DFID" (%p)", PFID(fid), inode);
191 if (bits & MDS_INODELOCK_OPEN) {
193 switch (lock->l_req_mode) {
204 CERROR("Unexpected lock mode for OPEN lock "
205 "%d, inode %ld\n", lock->l_req_mode,
208 ll_md_real_close(inode, flags);
211 if (bits & MDS_INODELOCK_UPDATE)
212 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
214 if (S_ISDIR(inode->i_mode) &&
215 (bits & MDS_INODELOCK_UPDATE)) {
216 CDEBUG(D_INODE, "invalidating inode %lu\n",
218 truncate_inode_pages(inode->i_mapping, 0);
219 ll_drop_negative_dentry(inode);
222 if (inode->i_sb->s_root &&
223 inode != inode->i_sb->s_root->d_inode &&
224 (bits & MDS_INODELOCK_LOOKUP))
225 ll_unhash_aliases(inode);
236 /* Pack the required supplementary groups into the supplied groups array.
237 * If we don't need to use the groups from the target inode(s) then we
238 * instead pack one or more groups from the user's supplementary group
239 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
240 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
245 LASSERT(suppgids != NULL);
247 if (in_group_p(i1->i_gid))
248 suppgids[0] = i1->i_gid;
253 if (in_group_p(i2->i_gid))
254 suppgids[1] = i2->i_gid;
261 for (i = 0; i < current_ngroups; i++) {
262 if (suppgids[0] == -1) {
263 if (current_groups[i] != suppgids[1])
264 suppgids[0] = current_groups[i];
267 if (suppgids[1] == -1) {
268 if (current_groups[i] != suppgids[0])
269 suppgids[1] = current_groups[i];
276 static void ll_d_add(struct dentry *de, struct inode *inode)
278 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
280 if (!list_empty(&de->d_alias)) {
281 spin_unlock(&dcache_lock);
282 CERROR("dentry %.*s %p alias next %p, prev %p\n",
283 de->d_name.len, de->d_name.name, de,
284 de->d_alias.next, de->d_alias.prev);
288 list_add(&de->d_alias, &inode->i_dentry);
292 if (!d_unhashed(de)) {
293 spin_unlock(&dcache_lock);
294 CERROR("dentry %.*s %p hash next %p\n",
295 de->d_name.len, de->d_name.name, de, de->d_hash.next);
301 /* Search "inode"'s alias list for a dentry that has the same name and parent
302 * as de. If found, return it. If not found, return de.
303 * Lustre can't use d_add_unique because don't unhash aliases for directory
304 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
305 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
307 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
309 struct list_head *tmp;
310 struct dentry *dentry;
311 struct dentry *last_discon = NULL;
313 spin_lock(&dcache_lock);
314 list_for_each(tmp, &inode->i_dentry) {
315 dentry = list_entry(tmp, struct dentry, d_alias);
317 /* We are called here with 'de' already on the aliases list. */
318 if (unlikely(dentry == de)) {
323 if (dentry->d_flags & DCACHE_DISCONNECTED) {
324 LASSERT(last_discon == NULL);
325 last_discon = dentry;
329 if (dentry->d_parent != de->d_parent)
332 if (dentry->d_name.hash != de->d_name.hash)
335 if (dentry->d_name.len != de->d_name.len)
338 if (memcmp(dentry->d_name.name, de->d_name.name,
339 de->d_name.len) != 0)
345 #ifdef LUSTRE_KERNEL_VERSION
346 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
348 unlock_dentry(dentry);
349 __d_rehash(dentry, 0); /* avoid taking dcache_lock inside */
350 spin_unlock(&dcache_lock);
352 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
353 "refc %d\n", de->d_name.len, de->d_name.name, de,
354 de->d_parent, de->d_inode, atomic_read(&de->d_count));
359 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
360 "refc %d\n", last_discon, last_discon->d_inode,
361 atomic_read(&last_discon->d_count));
363 dget_locked(last_discon);
364 __d_move(last_discon, de);
365 spin_unlock(&dcache_lock);
373 spin_unlock(&dcache_lock);
378 static int lookup_it_finish(struct ptlrpc_request *request, int offset,
379 struct lookup_intent *it, void *data)
381 struct it_cb_data *icbd = data;
382 struct dentry **de = icbd->icbd_childp;
383 struct inode *parent = icbd->icbd_parent;
384 struct ll_sb_info *sbi = ll_i2sbi(parent);
385 struct inode *inode = NULL;
388 /* NB 1 request reference will be taken away by ll_intent_lock()
390 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
393 rc = ll_prep_inode(&inode, request, offset,
398 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
399 inode, inode->i_ino, inode->i_generation);
400 md_set_lock_data(sbi->ll_md_exp,
401 &it->d.lustre.it_lock_handle, inode);
403 /* We used to query real size from OSTs here, but actually
404 this is not needed. For stat() calls size would be updated
405 from subsequent do_revalidate()->ll_inode_revalidate_it() in
407 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
408 Everybody else who needs correct file size would call
409 ll_glimpse_size or some equivalent themselves anyway.
410 Also see bug 7198. */
412 *de = ll_find_alias(inode, *de);
415 /* Check that parent has UPDATE lock. If there is none, we
416 cannot afford to hash this dentry (done by ll_d_add) as it
417 might get picked up later when UPDATE lock will appear */
418 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
419 spin_lock(&dcache_lock);
420 ll_d_add(*de, inode);
421 spin_unlock(&dcache_lock);
423 (*de)->d_inode = NULL;
424 /* We do not want to hash the dentry if don`t have a
425 * lock, but if this dentry is later used in d_move,
426 * we'd hit uninitialised list head d_hash, so we just
427 * do this to init d_hash field but leave dentry
428 * unhashed. (bug 10796). */
435 (*de)->d_op = &ll_d_ops;
440 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
441 struct lookup_intent *it, int lookup_flags)
443 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
444 struct dentry *save = dentry, *retval;
445 struct ptlrpc_request *req = NULL;
446 struct md_op_data *op_data;
447 struct it_cb_data icbd;
452 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
453 RETURN(ERR_PTR(-ENAMETOOLONG));
455 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
456 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
457 parent->i_generation, parent, LL_IT2STR(it));
459 if (d_mountpoint(dentry))
460 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
462 ll_frob_intent(&it, &lookup_it);
464 /* As do_lookup is called before follow_mount, root dentry may be left
465 * not valid, revalidate it here. */
466 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
467 (it->it_op & (IT_OPEN | IT_CREAT))) {
468 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
473 icbd.icbd_childp = &dentry;
474 icbd.icbd_parent = parent;
476 if (it->it_op & IT_CREAT ||
477 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
478 opc = LUSTRE_OPC_CREATE;
480 opc = LUSTRE_OPC_ANY;
482 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
483 dentry->d_name.len, lookup_flags, opc,
486 RETURN((void *)op_data);
488 it->it_create_mode &= ~current->fs->umask;
490 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
491 lookup_flags, &req, ll_md_blocking_ast, 0);
492 ll_finish_md_op_data(op_data);
494 GOTO(out, retval = ERR_PTR(rc));
496 rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, &icbd);
498 ll_intent_release(it);
499 GOTO(out, retval = ERR_PTR(rc));
502 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
503 !S_ISREG(dentry->d_inode->i_mode) &&
504 !S_ISDIR(dentry->d_inode->i_mode)) {
505 ll_release_openhandle(dentry, it);
507 ll_lookup_finish_locks(it, dentry);
510 GOTO(out, retval = NULL);
512 GOTO(out, retval = dentry);
515 ptlrpc_req_finished(req);
519 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
520 #ifdef LUSTRE_KERNEL_VERSION
521 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
522 struct nameidata *nd)
527 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
528 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
530 de = ll_lookup_it(parent, dentry, NULL, 0);
535 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
538 struct lookup_intent *it;
540 OBD_ALLOC(it, sizeof(*it));
542 return ERR_PTR(-ENOMEM);
544 if (lookup_flags & LOOKUP_OPEN) {
546 if (lookup_flags & LOOKUP_CREATE)
547 it->it_op |= IT_CREAT;
548 it->it_create_mode = oit->create_mode;
549 it->it_flags = oit->flags;
551 it->it_op = IT_GETATTR;
554 #ifndef HAVE_FILE_IN_STRUCT_INTENT
555 /* Since there is no way to pass our intent to ll_file_open,
556 * just check the file is there. Actual open will be done
558 if (it->it_op & IT_OPEN)
559 it->it_op = IT_LOOKUP;
565 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
566 struct nameidata *nd)
571 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
572 struct lookup_intent *it;
574 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
575 /* Did we came here from failed revalidate just to propagate
577 if (nd->flags & LOOKUP_OPEN)
578 if (IS_ERR(nd->intent.open.file))
579 RETURN((struct dentry *)nd->intent.open.file);
582 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
583 it = ll_d2d(dentry)->lld_it;
584 ll_d2d(dentry)->lld_it = NULL;
586 it = ll_convert_intent(&nd->intent.open, nd->flags);
588 RETURN((struct dentry *)it);
591 de = ll_lookup_it(parent, dentry, it, nd->flags);
594 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
595 if (dentry->d_inode &&
596 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
597 #ifdef HAVE_FILE_IN_STRUCT_INTENT
598 if (S_ISFIFO(dentry->d_inode->i_mode)) {
599 // We cannot call open here as it would
602 (struct ptlrpc_request *)
603 it->d.lustre.it_data);
606 nd->intent.open.file->private_data = it;
607 filp =lookup_instantiate_filp(nd,dentry,
609 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
610 /* 2.6.1[456] have a bug in open_namei() that forgets to check
611 * nd->intent.open.file for error, so we need to return it as lookup's result
616 de = (struct dentry *) filp;
621 #else /* HAVE_FILE_IN_STRUCT_INTENT */
622 /* Release open handle as we have no way to
623 * pass it to ll_file_open */
624 ll_release_openhandle(dentry, it);
625 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
626 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
627 // XXX This can only reliably work on assumption
628 // that there are NO hashed negative dentries.
629 ll_d2d(dentry)->lld_it = it;
630 it = NULL; /* Will be freed in ll_create_nd */
631 /* We absolutely depend on ll_create_nd to be
632 * called to not leak this intent and possible
633 * data attached to it */
638 ll_intent_release(it);
639 OBD_FREE(it, sizeof(*it));
642 de = ll_lookup_it(parent, dentry, NULL, 0);
650 /* We depend on "mode" being set with the proper file type/umask by now */
651 static struct inode *ll_create_node(struct inode *dir, const char *name,
652 int namelen, const void *data, int datalen,
653 int mode, __u64 extra,
654 struct lookup_intent *it)
656 struct inode *inode = NULL;
657 struct ptlrpc_request *request = NULL;
658 struct ll_sb_info *sbi = ll_i2sbi(dir);
662 LASSERT(it && it->d.lustre.it_disposition);
664 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
665 request = it->d.lustre.it_data;
666 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
667 rc = ll_prep_inode(&inode, request, DLM_REPLY_REC_OFF, dir->i_sb);
669 GOTO(out, inode = ERR_PTR(rc));
671 LASSERT(list_empty(&inode->i_dentry));
673 /* We asked for a lock on the directory, but were granted a
674 * lock on the inode. Since we finally have an inode pointer,
675 * stuff it in the lock. */
676 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
677 inode, inode->i_ino, inode->i_generation);
678 md_set_lock_data(sbi->ll_md_exp,
679 &it->d.lustre.it_lock_handle, inode);
682 ptlrpc_req_finished(request);
687 * By the time this is called, we already have created the directory cache
688 * entry for the new file, but it is so far negative - it has no inode.
690 * We defer creating the OBD object(s) until open, to keep the intent and
691 * non-intent code paths similar, and also because we do not have the MDS
692 * inode number before calling ll_create_node() (which is needed for LOV),
693 * so we would need to do yet another RPC to the MDS to store the LOV EA
694 * data on the MDS. If needed, we would pass the PACKED lmm as data and
695 * lmm_size in datalen (the MDS still has code which will handle that).
697 * If the create succeeds, we fill in the inode information
698 * with d_instantiate().
700 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
701 struct lookup_intent *it)
707 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
708 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
709 dir->i_generation, dir, LL_IT2STR(it));
711 rc = it_open_error(DISP_OPEN_CREATE, it);
715 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
716 NULL, 0, mode, 0, it);
718 RETURN(PTR_ERR(inode));
721 d_instantiate(dentry, inode);
722 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
723 * but some callers, e.g. do_coredump, expect dentry to be hashed after
724 * successful create. Hash it here. */
725 spin_lock(&dcache_lock);
726 if (d_unhashed(dentry))
727 __d_rehash(dentry, 0);
728 spin_unlock(&dcache_lock);
732 static void ll_update_times(struct ptlrpc_request *request, int offset,
735 struct mdt_body *body = lustre_msg_buf(request->rq_repmsg, offset,
739 /* mtime is always updated with ctime, but can be set in past.
740 As write and utime(2) may happen within 1 second, and utime's
741 mtime has a priority over write's one, so take mtime from mds
742 for the same ctimes. */
743 if (body->valid & OBD_MD_FLCTIME &&
744 body->ctime >= LTIME_S(inode->i_ctime)) {
745 LTIME_S(inode->i_ctime) = body->ctime;
747 if (body->valid & OBD_MD_FLMTIME) {
748 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
749 "to "LPU64"\n", inode->i_ino,
750 LTIME_S(inode->i_mtime), body->mtime);
751 LTIME_S(inode->i_mtime) = body->mtime;
756 static int ll_new_node(struct inode *dir, struct qstr *name,
757 const char *tgt, int mode, int rdev,
758 struct dentry *dchild, __u32 opc)
760 struct ptlrpc_request *request = NULL;
761 struct md_op_data *op_data;
762 struct inode *inode = NULL;
763 struct ll_sb_info *sbi = ll_i2sbi(dir);
768 if (unlikely(tgt != NULL))
769 tgt_len = strlen(tgt) + 1;
771 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
772 name->len, 0, opc, NULL);
774 GOTO(err_exit, err = PTR_ERR(op_data));
776 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
777 current->fsuid, current->fsgid,
778 current->cap_effective, rdev, &request);
779 ll_finish_md_op_data(op_data);
783 ll_update_times(request, REPLY_REC_OFF, dir);
786 err = ll_prep_inode(&inode, request, REPLY_REC_OFF,
792 d_instantiate(dchild, inode);
796 ptlrpc_req_finished(request);
801 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
802 unsigned rdev, struct dentry *dchild)
807 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
808 name->len, name->name, dir->i_ino, dir->i_generation, dir,
811 mode &= ~current->fs->umask;
813 switch (mode & S_IFMT) {
815 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
821 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
833 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
834 #ifndef LUSTRE_KERNEL_VERSION
835 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
836 int mode, struct nameidata *nd)
838 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
842 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
844 ll_d2d(dentry)->lld_it = NULL;
846 /* Was there an error? Propagate it! */
847 if (it->d.lustre.it_status) {
848 rc = it->d.lustre.it_status;
852 rc = ll_create_it(dir, dentry, mode, it);
853 #ifdef HAVE_FILE_IN_STRUCT_INTENT
854 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
855 nd->intent.open.file->private_data = it;
856 lookup_instantiate_filp(nd, dentry, NULL);
859 ll_release_openhandle(dentry,it);
863 ll_intent_release(it);
864 OBD_FREE(it, sizeof(*it));
869 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
870 int mode, struct nameidata *nd)
872 if (!nd || !nd->intent.d.lustre.it_disposition)
873 /* No saved request? Just mknod the file */
874 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
876 return ll_create_it(dir, dentry, mode, &nd->intent);
881 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
882 const char *tgt, struct dentry *dchild)
887 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
888 name->len, name->name, dir->i_ino, dir->i_generation,
891 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
892 0, dchild, LUSTRE_OPC_SYMLINK);
896 static int ll_link_generic(struct inode *src, struct inode *dir,
897 struct qstr *name, struct dentry *dchild)
899 struct ll_sb_info *sbi = ll_i2sbi(dir);
900 struct ptlrpc_request *request = NULL;
901 struct md_op_data *op_data;
906 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
907 src->i_ino, src->i_generation, src, dir->i_ino,
908 dir->i_generation, dir, name->len, name->name);
910 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
911 0, LUSTRE_OPC_ANY, NULL);
913 RETURN(PTR_ERR(op_data));
915 err = md_link(sbi->ll_md_exp, op_data, &request);
916 ll_finish_md_op_data(op_data);
922 ll_update_times(request, REPLY_REC_OFF, dir);
925 ptlrpc_req_finished(request);
929 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
930 int mode, struct dentry *dchild)
936 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
937 name->len, name->name, dir->i_ino, dir->i_generation, dir);
939 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
940 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
945 /* Try to find the child dentry by its name.
946 If found, put the result fid into @fid. */
947 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
950 struct dentry *parent, *child;
952 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
953 child = d_lookup(parent, name);
956 *fid = *ll_inode2fid(child->d_inode);
961 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
962 struct dentry *dchild, struct qstr *name)
964 struct ptlrpc_request *request = NULL;
965 struct md_op_data *op_data;
969 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
970 name->len, name->name, dir->i_ino, dir->i_generation, dir);
972 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
975 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
976 S_IFDIR, LUSTRE_OPC_ANY, NULL);
978 RETURN(PTR_ERR(op_data));
980 ll_get_child_fid(dir, name, &op_data->op_fid3);
981 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
982 ll_finish_md_op_data(op_data);
984 ll_update_times(request, REPLY_REC_OFF, dir);
985 ptlrpc_req_finished(request);
989 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
991 struct mdt_body *body;
992 struct lov_mds_md *eadata;
993 struct lov_stripe_md *lsm = NULL;
994 struct obd_trans_info oti = { 0 };
999 /* req is swabbed so this is safe */
1000 body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
1002 if (!(body->valid & OBD_MD_FLEASIZE))
1005 if (body->eadatasize == 0) {
1006 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1007 GOTO(out, rc = -EPROTO);
1010 /* The MDS sent back the EA because we unlinked the last reference
1011 * to this file. Use this EA to unlink the objects on the OST.
1012 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1013 * check it is complete and sensible. */
1014 eadata = lustre_swab_repbuf(request, REPLY_REC_OFF + 1,
1015 body->eadatasize, NULL);
1016 LASSERT(eadata != NULL);
1017 if (eadata == NULL) {
1018 CERROR("Can't unpack MDS EA data\n");
1019 GOTO(out, rc = -EPROTO);
1022 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1024 CERROR("obd_unpackmd: %d\n", rc);
1027 LASSERT(rc >= sizeof(*lsm));
1029 rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1031 GOTO(out_free_memmd, rc);
1035 GOTO(out_free_memmd, rc = -ENOMEM);
1037 oa->o_id = lsm->lsm_object_id;
1038 oa->o_gr = lsm->lsm_object_gr;
1039 oa->o_mode = body->mode & S_IFMT;
1040 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1042 if (body->valid & OBD_MD_FLCOOKIE) {
1043 oa->o_valid |= OBD_MD_FLCOOKIE;
1044 oti.oti_logcookies =
1045 lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF + 2,
1046 sizeof(struct llog_cookie) *
1047 lsm->lsm_stripe_count);
1048 if (oti.oti_logcookies == NULL) {
1049 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1050 body->valid &= ~OBD_MD_FLCOOKIE;
1054 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir));
1057 CERROR("obd destroy objid "LPX64" error %d\n",
1058 lsm->lsm_object_id, rc);
1060 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1065 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1066 struct dentry *dchild, struct qstr *name)
1068 struct ptlrpc_request *request = NULL;
1069 struct md_op_data *op_data;
1072 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1073 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1076 * XXX: unlink bind mountpoint maybe call to here,
1077 * just check it as vfs_unlink does.
1079 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1082 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1083 name->len, 0, LUSTRE_OPC_ANY, NULL);
1084 if (IS_ERR(op_data))
1085 RETURN(PTR_ERR(op_data));
1087 ll_get_child_fid(dir, name, &op_data->op_fid3);
1088 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1089 ll_finish_md_op_data(op_data);
1094 ll_update_times(request, REPLY_REC_OFF, dir);
1096 rc = ll_objects_destroy(request, dir);
1098 ptlrpc_req_finished(request);
1102 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1103 struct dentry *src_dchild, struct qstr *src_name,
1104 struct inode *tgt, struct dentry *tgt_dparent,
1105 struct dentry *tgt_dchild, struct qstr *tgt_name)
1107 struct ptlrpc_request *request = NULL;
1108 struct ll_sb_info *sbi = ll_i2sbi(src);
1109 struct md_op_data *op_data;
1112 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1113 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1114 src->i_ino, src->i_generation, src, tgt_name->len,
1115 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1117 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1118 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1121 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1122 LUSTRE_OPC_ANY, NULL);
1123 if (IS_ERR(op_data))
1124 RETURN(PTR_ERR(op_data));
1126 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1127 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1128 err = md_rename(sbi->ll_md_exp, op_data,
1129 src_name->name, src_name->len,
1130 tgt_name->name, tgt_name->len, &request);
1131 ll_finish_md_op_data(op_data);
1133 ll_update_times(request, REPLY_REC_OFF, src);
1134 ll_update_times(request, REPLY_REC_OFF, tgt);
1135 err = ll_objects_destroy(request, src);
1138 ptlrpc_req_finished(request);
1143 #ifdef LUSTRE_KERNEL_VERSION
1144 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1146 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1148 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1150 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1152 tgtnd->dentry->d_inode, tgtnd->dentry,
1153 NULL, &tgtnd->last);
1155 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1157 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1158 &tgtnd->last, NULL);
1160 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1162 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1164 static int ll_rmdir_raw(struct nameidata *nd)
1166 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1169 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1171 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1173 static int ll_unlink_raw(struct nameidata *nd)
1175 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1180 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1183 return ll_mknod_generic(dir, &dchild->d_name, mode,
1184 old_encode_dev(rdev), dchild);
1187 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1188 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1190 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1192 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1194 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1196 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1198 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1200 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1201 const char *oldname)
1203 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1205 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1206 struct dentry *new_dentry)
1208 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1211 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1212 struct inode *new_dir, struct dentry *new_dentry)
1214 return ll_rename_generic(old_dir, NULL,
1215 old_dentry, &old_dentry->d_name,
1216 new_dir, NULL, new_dentry,
1217 &new_dentry->d_name);
1221 struct inode_operations ll_dir_inode_operations = {
1222 #ifdef LUSTRE_KERNEL_VERSION
1223 .link_raw = ll_link_raw,
1224 .unlink_raw = ll_unlink_raw,
1225 .symlink_raw = ll_symlink_raw,
1226 .mkdir_raw = ll_mkdir_raw,
1227 .rmdir_raw = ll_rmdir_raw,
1228 .mknod_raw = ll_mknod_raw,
1229 .rename_raw = ll_rename_raw,
1230 .setattr = ll_setattr,
1231 .setattr_raw = ll_setattr_raw,
1234 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1235 .create_it = ll_create_it,
1236 .lookup_it = ll_lookup_it,
1237 .revalidate_it = ll_inode_revalidate_it,
1239 .lookup = ll_lookup_nd,
1240 .create = ll_create_nd,
1241 /* We need all these non-raw things for NFSD, to not patch it. */
1242 .unlink = ll_unlink,
1245 .symlink = ll_symlink,
1247 .rename = ll_rename,
1248 .setattr = ll_setattr,
1249 .getattr = ll_getattr,
1251 .permission = ll_inode_permission,
1252 .setxattr = ll_setxattr,
1253 .getxattr = ll_getxattr,
1254 .listxattr = ll_listxattr,
1255 .removexattr = ll_removexattr,
1258 struct inode_operations ll_special_inode_operations = {
1259 #ifdef LUSTRE_KERNEL_VERSION
1260 .setattr_raw = ll_setattr_raw,
1262 .setattr = ll_setattr,
1263 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1264 .getattr = ll_getattr,
1266 .revalidate_it = ll_inode_revalidate_it,
1268 .permission = ll_inode_permission,
1269 .setxattr = ll_setxattr,
1270 .getxattr = ll_getxattr,
1271 .listxattr = ll_listxattr,
1272 .removexattr = ll_removexattr,