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 struct inode *ll_iget(struct super_block *sb, ino_t hash,
80 struct ll_inode_info *lli;
84 inode = iget_locked(sb, hash);
86 if (inode->i_state & I_NEW) {
87 lli = ll_i2info(inode);
88 ll_read_inode2(inode, md);
89 unlock_new_inode(inode);
91 if (!(inode->i_state & (I_FREEING | I_CLEAR)))
92 ll_update_inode(inode, md);
94 CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n",
95 inode->i_ino, inode->i_generation, inode);
101 static void ll_drop_negative_dentry(struct inode *dir)
103 struct dentry *dentry, *tmp_alias, *tmp_subdir;
105 spin_lock(&dcache_lock);
107 list_for_each_entry_safe(dentry, tmp_alias,
108 &dir->i_dentry,d_alias) {
109 if (!list_empty(&dentry->d_subdirs)) {
110 struct dentry *child;
111 list_for_each_entry_safe(child, tmp_subdir,
114 /* XXX Print some debug here? */
116 /* Negative dentry. If we were
117 dropping dcache lock, go
118 throught the list again */
119 if (ll_drop_dentry(child))
124 spin_unlock(&dcache_lock);
128 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
129 void *data, int flag)
132 struct lustre_handle lockh;
136 case LDLM_CB_BLOCKING:
137 ldlm_lock2handle(lock, &lockh);
138 rc = ldlm_cli_cancel(&lockh);
140 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
144 case LDLM_CB_CANCELING: {
145 struct inode *inode = ll_inode_from_lock(lock);
146 __u64 bits = lock->l_policy_data.l_inodebits.bits;
149 /* Invalidate all dentries associated with this inode */
153 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
154 if ((bits & MDS_INODELOCK_LOOKUP) &&
155 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
156 bits &= ~MDS_INODELOCK_LOOKUP;
157 if ((bits & MDS_INODELOCK_UPDATE) &&
158 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
159 bits &= ~MDS_INODELOCK_UPDATE;
160 if ((bits & MDS_INODELOCK_OPEN) &&
161 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
162 bits &= ~MDS_INODELOCK_OPEN;
164 fid = ll_inode2fid(inode);
165 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
166 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
167 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
168 LDLM_ERROR(lock, "data mismatch with object "
169 DFID" (%p)", PFID(fid), inode);
172 if (bits & MDS_INODELOCK_OPEN) {
174 switch (lock->l_req_mode) {
185 CERROR("Unexpected lock mode for OPEN lock "
186 "%d, inode %ld\n", lock->l_req_mode,
189 ll_md_real_close(inode, flags);
192 if (bits & MDS_INODELOCK_UPDATE)
193 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
195 if (S_ISDIR(inode->i_mode) &&
196 (bits & MDS_INODELOCK_UPDATE)) {
197 CDEBUG(D_INODE, "invalidating inode %lu\n",
199 truncate_inode_pages(inode->i_mapping, 0);
200 ll_drop_negative_dentry(inode);
203 if (inode->i_sb->s_root &&
204 inode != inode->i_sb->s_root->d_inode &&
205 (bits & MDS_INODELOCK_LOOKUP))
206 ll_unhash_aliases(inode);
217 /* Pack the required supplementary groups into the supplied groups array.
218 * If we don't need to use the groups from the target inode(s) then we
219 * instead pack one or more groups from the user's supplementary group
220 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
221 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
226 LASSERT(suppgids != NULL);
228 if (in_group_p(i1->i_gid))
229 suppgids[0] = i1->i_gid;
234 if (in_group_p(i2->i_gid))
235 suppgids[1] = i2->i_gid;
242 for (i = 0; i < current_ngroups; i++) {
243 if (suppgids[0] == -1) {
244 if (current_groups[i] != suppgids[1])
245 suppgids[0] = current_groups[i];
248 if (suppgids[1] == -1) {
249 if (current_groups[i] != suppgids[0])
250 suppgids[1] = current_groups[i];
257 static void ll_d_add(struct dentry *de, struct inode *inode)
259 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
261 if (!list_empty(&de->d_alias)) {
262 spin_unlock(&dcache_lock);
263 CERROR("dentry %.*s %p alias next %p, prev %p\n",
264 de->d_name.len, de->d_name.name, de,
265 de->d_alias.next, de->d_alias.prev);
269 list_add(&de->d_alias, &inode->i_dentry);
273 if (!d_unhashed(de)) {
274 spin_unlock(&dcache_lock);
275 CERROR("dentry %.*s %p hash next %p\n",
276 de->d_name.len, de->d_name.name, de, de->d_hash.next);
282 /* Search "inode"'s alias list for a dentry that has the same name and parent
283 * as de. If found, return it. If not found, return de.
284 * Lustre can't use d_add_unique because don't unhash aliases for directory
285 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
286 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
288 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
290 struct list_head *tmp;
291 struct dentry *dentry;
292 struct dentry *last_discon = NULL;
294 spin_lock(&dcache_lock);
295 list_for_each(tmp, &inode->i_dentry) {
296 dentry = list_entry(tmp, struct dentry, d_alias);
298 /* We are called here with 'de' already on the aliases list. */
299 if (unlikely(dentry == de)) {
304 if (dentry->d_flags & DCACHE_DISCONNECTED) {
305 LASSERT(last_discon == NULL);
306 last_discon = dentry;
310 if (dentry->d_parent != de->d_parent)
313 if (dentry->d_name.hash != de->d_name.hash)
316 if (dentry->d_name.len != de->d_name.len)
319 if (memcmp(dentry->d_name.name, de->d_name.name,
320 de->d_name.len) != 0)
326 #ifdef DCACHE_LUSTRE_INVALID
327 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
329 unlock_dentry(dentry);
330 __d_rehash(dentry, 0); /* avoid taking dcache_lock inside */
331 spin_unlock(&dcache_lock);
333 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
334 "refc %d\n", de->d_name.len, de->d_name.name, de,
335 de->d_parent, de->d_inode, atomic_read(&de->d_count));
340 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
341 "refc %d\n", last_discon, last_discon->d_inode,
342 atomic_read(&last_discon->d_count));
343 dget_locked(last_discon);
344 spin_unlock(&dcache_lock);
346 d_move(last_discon, de);
353 spin_unlock(&dcache_lock);
358 static int lookup_it_finish(struct ptlrpc_request *request, int offset,
359 struct lookup_intent *it, void *data)
361 struct it_cb_data *icbd = data;
362 struct dentry **de = icbd->icbd_childp;
363 struct inode *parent = icbd->icbd_parent;
364 struct ll_sb_info *sbi = ll_i2sbi(parent);
365 struct inode *inode = NULL;
368 /* NB 1 request reference will be taken away by ll_intent_lock()
370 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
373 rc = ll_prep_inode(&inode, request, offset,
378 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
379 inode, inode->i_ino, inode->i_generation);
380 md_set_lock_data(sbi->ll_md_exp,
381 &it->d.lustre.it_lock_handle, inode);
383 /* We used to query real size from OSTs here, but actually
384 this is not needed. For stat() calls size would be updated
385 from subsequent do_revalidate()->ll_inode_revalidate_it() in
387 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
388 Everybody else who needs correct file size would call
389 ll_glimpse_size or some equivalent themselves anyway.
390 Also see bug 7198. */
392 *de = ll_find_alias(inode, *de);
395 /* Check that parent has UPDATE lock. If there is none, we
396 cannot afford to hash this dentry (done by ll_d_add) as it
397 might get picked up later when UPDATE lock will appear */
398 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
399 spin_lock(&dcache_lock);
400 ll_d_add(*de, inode);
401 spin_unlock(&dcache_lock);
403 (*de)->d_inode = NULL;
404 /* We do not want to hash the dentry if don`t have a
405 * lock, but if this dentry is later used in d_move,
406 * we'd hit uninitialised list head d_hash, so we just
407 * do this to init d_hash field but leave dentry
408 * unhashed. (bug 10796). */
415 (*de)->d_op = &ll_d_ops;
420 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
421 struct lookup_intent *it, int lookup_flags)
423 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
424 struct dentry *save = dentry, *retval;
425 struct ptlrpc_request *req = NULL;
426 struct md_op_data *op_data;
427 struct it_cb_data icbd;
432 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
433 RETURN(ERR_PTR(-ENAMETOOLONG));
435 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
436 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
437 parent->i_generation, parent, LL_IT2STR(it));
439 if (d_mountpoint(dentry))
440 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
442 ll_frob_intent(&it, &lookup_it);
444 /* As do_lookup is called before follow_mount, root dentry may be left
445 * not valid, revalidate it here. */
446 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
447 (it->it_op & (IT_OPEN | IT_CREAT))) {
448 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
453 icbd.icbd_childp = &dentry;
454 icbd.icbd_parent = parent;
456 if (it->it_op & IT_CREAT ||
457 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
458 opc = LUSTRE_OPC_CREATE;
460 opc = LUSTRE_OPC_ANY;
462 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
463 dentry->d_name.len, lookup_flags, opc,
466 RETURN((void *)op_data);
468 it->it_create_mode &= ~current->fs->umask;
470 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
471 lookup_flags, &req, ll_md_blocking_ast, 0);
472 ll_finish_md_op_data(op_data);
474 GOTO(out, retval = ERR_PTR(rc));
476 rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, &icbd);
478 ll_intent_release(it);
479 GOTO(out, retval = ERR_PTR(rc));
482 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
483 !S_ISREG(dentry->d_inode->i_mode) &&
484 !S_ISDIR(dentry->d_inode->i_mode)) {
485 ll_release_openhandle(dentry, it);
487 ll_lookup_finish_locks(it, dentry);
490 GOTO(out, retval = NULL);
492 GOTO(out, retval = dentry);
495 ptlrpc_req_finished(req);
499 #ifdef HAVE_VFS_INTENT_PATCHES
500 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
501 struct nameidata *nd)
506 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
507 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
509 de = ll_lookup_it(parent, dentry, NULL, 0);
514 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
517 struct lookup_intent *it;
519 OBD_ALLOC(it, sizeof(*it));
521 return ERR_PTR(-ENOMEM);
523 if (lookup_flags & LOOKUP_OPEN) {
525 if (lookup_flags & LOOKUP_CREATE)
526 it->it_op |= IT_CREAT;
527 it->it_create_mode = oit->create_mode;
528 it->it_flags = oit->flags;
530 it->it_op = IT_GETATTR;
533 #ifndef HAVE_FILE_IN_STRUCT_INTENT
534 /* Since there is no way to pass our intent to ll_file_open,
535 * just check the file is there. Actual open will be done
537 if (it->it_op & IT_OPEN)
538 it->it_op = IT_LOOKUP;
544 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
545 struct nameidata *nd)
550 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
551 struct lookup_intent *it;
553 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
554 /* Did we came here from failed revalidate just to propagate
556 if (nd->flags & LOOKUP_OPEN)
557 if (IS_ERR(nd->intent.open.file))
558 RETURN((struct dentry *)nd->intent.open.file);
561 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
562 it = ll_d2d(dentry)->lld_it;
563 ll_d2d(dentry)->lld_it = NULL;
565 it = ll_convert_intent(&nd->intent.open, nd->flags);
567 RETURN((struct dentry *)it);
570 de = ll_lookup_it(parent, dentry, it, nd->flags);
573 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
574 if (dentry->d_inode &&
575 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
576 #ifdef HAVE_FILE_IN_STRUCT_INTENT
577 if (S_ISFIFO(dentry->d_inode->i_mode)) {
578 // We cannot call open here as it would
581 (struct ptlrpc_request *)
582 it->d.lustre.it_data);
585 nd->intent.open.file->private_data = it;
586 filp =lookup_instantiate_filp(nd,dentry,
588 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
589 /* 2.6.1[456] have a bug in open_namei() that forgets to check
590 * nd->intent.open.file for error, so we need to return it as lookup's result
595 de = (struct dentry *) filp;
600 #else /* HAVE_FILE_IN_STRUCT_INTENT */
601 /* Release open handle as we have no way to
602 * pass it to ll_file_open */
603 ll_release_openhandle(dentry, it);
604 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
605 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
606 // XXX This can only reliably work on assumption
607 // that there are NO hashed negative dentries.
608 ll_d2d(dentry)->lld_it = it;
609 it = NULL; /* Will be freed in ll_create_nd */
610 /* We absolutely depend on ll_create_nd to be
611 * called to not leak this intent and possible
612 * data attached to it */
617 ll_intent_release(it);
618 OBD_FREE(it, sizeof(*it));
621 de = ll_lookup_it(parent, dentry, NULL, 0);
628 /* We depend on "mode" being set with the proper file type/umask by now */
629 static struct inode *ll_create_node(struct inode *dir, const char *name,
630 int namelen, const void *data, int datalen,
631 int mode, __u64 extra,
632 struct lookup_intent *it)
634 struct inode *inode = NULL;
635 struct ptlrpc_request *request = NULL;
636 struct ll_sb_info *sbi = ll_i2sbi(dir);
640 LASSERT(it && it->d.lustre.it_disposition);
642 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
643 request = it->d.lustre.it_data;
644 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
645 rc = ll_prep_inode(&inode, request, DLM_REPLY_REC_OFF, dir->i_sb);
647 GOTO(out, inode = ERR_PTR(rc));
649 LASSERT(list_empty(&inode->i_dentry));
651 /* We asked for a lock on the directory, but were granted a
652 * lock on the inode. Since we finally have an inode pointer,
653 * stuff it in the lock. */
654 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
655 inode, inode->i_ino, inode->i_generation);
656 md_set_lock_data(sbi->ll_md_exp,
657 &it->d.lustre.it_lock_handle, inode);
660 ptlrpc_req_finished(request);
665 * By the time this is called, we already have created the directory cache
666 * entry for the new file, but it is so far negative - it has no inode.
668 * We defer creating the OBD object(s) until open, to keep the intent and
669 * non-intent code paths similar, and also because we do not have the MDS
670 * inode number before calling ll_create_node() (which is needed for LOV),
671 * so we would need to do yet another RPC to the MDS to store the LOV EA
672 * data on the MDS. If needed, we would pass the PACKED lmm as data and
673 * lmm_size in datalen (the MDS still has code which will handle that).
675 * If the create succeeds, we fill in the inode information
676 * with d_instantiate().
678 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
679 struct lookup_intent *it)
685 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
686 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
687 dir->i_generation, dir, LL_IT2STR(it));
689 rc = it_open_error(DISP_OPEN_CREATE, it);
693 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
694 NULL, 0, mode, 0, it);
696 RETURN(PTR_ERR(inode));
699 d_instantiate(dentry, inode);
700 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
701 * but some callers, e.g. do_coredump, expect dentry to be hashed after
702 * successful create. Hash it here. */
703 spin_lock(&dcache_lock);
704 if (d_unhashed(dentry))
705 __d_rehash(dentry, 0);
706 spin_unlock(&dcache_lock);
710 static void ll_update_times(struct ptlrpc_request *request, int offset,
713 struct mdt_body *body = lustre_msg_buf(request->rq_repmsg, offset,
717 /* mtime is always updated with ctime, but can be set in past.
718 As write and utime(2) may happen within 1 second, and utime's
719 mtime has a priority over write's one, so take mtime from mds
720 for the same ctimes. */
721 if (body->valid & OBD_MD_FLCTIME &&
722 body->ctime >= LTIME_S(inode->i_ctime)) {
723 LTIME_S(inode->i_ctime) = body->ctime;
725 if (body->valid & OBD_MD_FLMTIME) {
726 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
727 "to "LPU64"\n", inode->i_ino,
728 LTIME_S(inode->i_mtime), body->mtime);
729 LTIME_S(inode->i_mtime) = body->mtime;
734 static int ll_new_node(struct inode *dir, struct qstr *name,
735 const char *tgt, int mode, int rdev,
736 struct dentry *dchild, __u32 opc)
738 struct ptlrpc_request *request = NULL;
739 struct md_op_data *op_data;
740 struct inode *inode = NULL;
741 struct ll_sb_info *sbi = ll_i2sbi(dir);
746 if (unlikely(tgt != NULL))
747 tgt_len = strlen(tgt) + 1;
749 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
750 name->len, 0, opc, NULL);
752 GOTO(err_exit, err = PTR_ERR(op_data));
754 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
755 current->fsuid, current->fsgid,
756 current->cap_effective, rdev, &request);
757 ll_finish_md_op_data(op_data);
761 ll_update_times(request, REPLY_REC_OFF, dir);
764 err = ll_prep_inode(&inode, request, REPLY_REC_OFF,
770 d_instantiate(dchild, inode);
774 ptlrpc_req_finished(request);
779 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
780 unsigned rdev, struct dentry *dchild)
785 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
786 name->len, name->name, dir->i_ino, dir->i_generation, dir,
789 mode &= ~current->fs->umask;
791 switch (mode & S_IFMT) {
793 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
799 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
811 #ifndef HAVE_VFS_INTENT_PATCHES
812 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
813 int mode, struct nameidata *nd)
815 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
819 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
821 ll_d2d(dentry)->lld_it = NULL;
823 /* Was there an error? Propagate it! */
824 if (it->d.lustre.it_status) {
825 rc = it->d.lustre.it_status;
829 rc = ll_create_it(dir, dentry, mode, it);
830 #ifdef HAVE_FILE_IN_STRUCT_INTENT
831 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
832 nd->intent.open.file->private_data = it;
833 lookup_instantiate_filp(nd, dentry, NULL);
836 ll_release_openhandle(dentry,it);
840 ll_intent_release(it);
841 OBD_FREE(it, sizeof(*it));
846 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
847 int mode, struct nameidata *nd)
849 if (!nd || !nd->intent.d.lustre.it_disposition)
850 /* No saved request? Just mknod the file */
851 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
853 return ll_create_it(dir, dentry, mode, &nd->intent);
857 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
858 const char *tgt, struct dentry *dchild)
863 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
864 name->len, name->name, dir->i_ino, dir->i_generation,
867 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
868 0, dchild, LUSTRE_OPC_SYMLINK);
872 static int ll_link_generic(struct inode *src, struct inode *dir,
873 struct qstr *name, struct dentry *dchild)
875 struct ll_sb_info *sbi = ll_i2sbi(dir);
876 struct ptlrpc_request *request = NULL;
877 struct md_op_data *op_data;
882 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
883 src->i_ino, src->i_generation, src, dir->i_ino,
884 dir->i_generation, dir, name->len, name->name);
886 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
887 0, LUSTRE_OPC_ANY, NULL);
889 RETURN(PTR_ERR(op_data));
891 err = md_link(sbi->ll_md_exp, op_data, &request);
892 ll_finish_md_op_data(op_data);
898 ll_update_times(request, REPLY_REC_OFF, dir);
901 ptlrpc_req_finished(request);
905 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
906 int mode, struct dentry *dchild)
912 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
913 name->len, name->name, dir->i_ino, dir->i_generation, dir);
915 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
916 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
921 /* Try to find the child dentry by its name.
922 If found, put the result fid into @fid. */
923 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
926 struct dentry *parent, *child;
928 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
929 child = d_lookup(parent, name);
932 *fid = *ll_inode2fid(child->d_inode);
937 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
938 struct dentry *dchild, struct qstr *name)
940 struct ptlrpc_request *request = NULL;
941 struct md_op_data *op_data;
945 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
946 name->len, name->name, dir->i_ino, dir->i_generation, dir);
948 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
951 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
952 S_IFDIR, LUSTRE_OPC_ANY, NULL);
954 RETURN(PTR_ERR(op_data));
956 ll_get_child_fid(dir, name, &op_data->op_fid3);
957 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
958 ll_finish_md_op_data(op_data);
960 ll_update_times(request, REPLY_REC_OFF, dir);
961 ptlrpc_req_finished(request);
965 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
967 struct mdt_body *body;
968 struct lov_mds_md *eadata;
969 struct lov_stripe_md *lsm = NULL;
970 struct obd_trans_info oti = { 0 };
975 /* req is swabbed so this is safe */
976 body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
978 if (!(body->valid & OBD_MD_FLEASIZE))
981 if (body->eadatasize == 0) {
982 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
983 GOTO(out, rc = -EPROTO);
986 /* The MDS sent back the EA because we unlinked the last reference
987 * to this file. Use this EA to unlink the objects on the OST.
988 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
989 * check it is complete and sensible. */
990 eadata = lustre_swab_repbuf(request, REPLY_REC_OFF + 1,
991 body->eadatasize, NULL);
992 LASSERT(eadata != NULL);
993 if (eadata == NULL) {
994 CERROR("Can't unpack MDS EA data\n");
995 GOTO(out, rc = -EPROTO);
998 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1000 CERROR("obd_unpackmd: %d\n", rc);
1003 LASSERT(rc >= sizeof(*lsm));
1005 rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1007 GOTO(out_free_memmd, rc);
1011 GOTO(out_free_memmd, rc = -ENOMEM);
1013 oa->o_id = lsm->lsm_object_id;
1014 oa->o_gr = lsm->lsm_object_gr;
1015 oa->o_mode = body->mode & S_IFMT;
1016 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1018 if (body->valid & OBD_MD_FLCOOKIE) {
1019 oa->o_valid |= OBD_MD_FLCOOKIE;
1020 oti.oti_logcookies =
1021 lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF + 2,
1022 sizeof(struct llog_cookie) *
1023 lsm->lsm_stripe_count);
1024 if (oti.oti_logcookies == NULL) {
1025 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1026 body->valid &= ~OBD_MD_FLCOOKIE;
1030 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir));
1033 CERROR("obd destroy objid "LPX64" error %d\n",
1034 lsm->lsm_object_id, rc);
1036 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1041 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1042 struct dentry *dchild, struct qstr *name)
1044 struct ptlrpc_request *request = NULL;
1045 struct md_op_data *op_data;
1048 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1049 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1052 * XXX: unlink bind mountpoint maybe call to here,
1053 * just check it as vfs_unlink does.
1055 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1058 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1059 name->len, 0, LUSTRE_OPC_ANY, NULL);
1060 if (IS_ERR(op_data))
1061 RETURN(PTR_ERR(op_data));
1063 ll_get_child_fid(dir, name, &op_data->op_fid3);
1064 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1065 ll_finish_md_op_data(op_data);
1069 ll_update_times(request, REPLY_REC_OFF, dir);
1071 rc = ll_objects_destroy(request, dir);
1073 ptlrpc_req_finished(request);
1077 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1078 struct dentry *src_dchild, struct qstr *src_name,
1079 struct inode *tgt, struct dentry *tgt_dparent,
1080 struct dentry *tgt_dchild, struct qstr *tgt_name)
1082 struct ptlrpc_request *request = NULL;
1083 struct ll_sb_info *sbi = ll_i2sbi(src);
1084 struct md_op_data *op_data;
1087 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1088 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1089 src->i_ino, src->i_generation, src, tgt_name->len,
1090 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1092 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1093 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1096 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1097 LUSTRE_OPC_ANY, NULL);
1098 if (IS_ERR(op_data))
1099 RETURN(PTR_ERR(op_data));
1101 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1102 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1103 err = md_rename(sbi->ll_md_exp, op_data,
1104 src_name->name, src_name->len,
1105 tgt_name->name, tgt_name->len, &request);
1106 ll_finish_md_op_data(op_data);
1108 ll_update_times(request, REPLY_REC_OFF, src);
1109 ll_update_times(request, REPLY_REC_OFF, tgt);
1110 err = ll_objects_destroy(request, src);
1113 ptlrpc_req_finished(request);
1118 #ifdef HAVE_VFS_INTENT_PATCHES
1119 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1121 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1123 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1125 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1127 tgtnd->dentry->d_inode, tgtnd->dentry,
1128 NULL, &tgtnd->last);
1130 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1132 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1133 &tgtnd->last, NULL);
1135 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1137 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1139 static int ll_rmdir_raw(struct nameidata *nd)
1141 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1144 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1146 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1148 static int ll_unlink_raw(struct nameidata *nd)
1150 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1155 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1158 return ll_mknod_generic(dir, &dchild->d_name, mode,
1159 old_encode_dev(rdev), dchild);
1162 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1164 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1166 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1168 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1170 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1172 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1174 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1175 const char *oldname)
1177 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1179 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1180 struct dentry *new_dentry)
1182 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1185 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1186 struct inode *new_dir, struct dentry *new_dentry)
1188 return ll_rename_generic(old_dir, NULL,
1189 old_dentry, &old_dentry->d_name,
1190 new_dir, NULL, new_dentry,
1191 &new_dentry->d_name);
1194 struct inode_operations ll_dir_inode_operations = {
1195 #ifdef HAVE_VFS_INTENT_PATCHES
1196 .link_raw = ll_link_raw,
1197 .unlink_raw = ll_unlink_raw,
1198 .symlink_raw = ll_symlink_raw,
1199 .mkdir_raw = ll_mkdir_raw,
1200 .rmdir_raw = ll_rmdir_raw,
1201 .mknod_raw = ll_mknod_raw,
1202 .rename_raw = ll_rename_raw,
1203 .setattr = ll_setattr,
1204 .setattr_raw = ll_setattr_raw,
1207 .lookup = ll_lookup_nd,
1208 .create = ll_create_nd,
1209 /* We need all these non-raw things for NFSD, to not patch it. */
1210 .unlink = ll_unlink,
1213 .symlink = ll_symlink,
1215 .rename = ll_rename,
1216 .setattr = ll_setattr,
1217 .getattr = ll_getattr,
1218 .permission = ll_inode_permission,
1219 .setxattr = ll_setxattr,
1220 .getxattr = ll_getxattr,
1221 .listxattr = ll_listxattr,
1222 .removexattr = ll_removexattr,
1225 struct inode_operations ll_special_inode_operations = {
1226 #ifdef HAVE_VFS_INTENT_PATCHES
1227 .setattr_raw = ll_setattr_raw,
1229 .setattr = ll_setattr,
1230 .getattr = ll_getattr,
1231 .permission = ll_inode_permission,
1232 .setxattr = ll_setxattr,
1233 .getxattr = ll_getxattr,
1234 .listxattr = ll_listxattr,
1235 .removexattr = ll_removexattr,