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 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 #include <linux/sched.h>
40 #include <linux/smp_lock.h>
41 #include <linux/quotaops.h>
42 #include <linux/highmem.h>
43 #include <linux/pagemap.h>
45 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <lustre_lite.h>
50 #include <lustre_dlm.h>
51 #include <lustre_ver.h>
52 #include <lustre_mdc.h>
53 #include "llite_internal.h"
56 * Check if we have something mounted at the named dchild.
57 * In such a case there would always be dentry present.
59 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
64 if (unlikely(dchild)) {
65 mounted = d_mountpoint(dchild);
67 dchild = d_lookup(dparent, name);
69 mounted = d_mountpoint(dchild);
76 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
80 ldlm_lock_decref(lockh, mode);
86 /* called from iget5_locked->find_inode() under inode_lock spinlock */
87 static int ll_test_inode(struct inode *inode, void *opaque)
89 struct ll_inode_info *lli = ll_i2info(inode);
90 struct lustre_md *md = opaque;
92 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
93 CERROR("MDS body missing FID\n");
97 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
103 static int ll_set_inode(struct inode *inode, void *opaque)
110 * Get an inode by inode number (already instantiated by the intent lookup).
111 * Returns inode or NULL
113 struct inode *ll_iget(struct super_block *sb, ino_t hash,
114 struct lustre_md *md)
116 struct ll_inode_info *lli;
121 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
124 lli = ll_i2info(inode);
125 if (inode->i_state & I_NEW) {
128 ll_read_inode2(inode, md);
129 rc = cl_inode_init(inode, md);
132 make_bad_inode(inode);
133 unlock_new_inode(inode);
137 unlock_new_inode(inode);
138 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
139 ll_update_inode(inode, md);
140 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
141 inode, PFID(&md->body->fid1));
146 static void ll_drop_negative_dentry(struct inode *dir)
148 struct dentry *dentry, *tmp_alias, *tmp_subdir;
150 spin_lock(&ll_lookup_lock);
151 spin_lock(&dcache_lock);
153 list_for_each_entry_safe(dentry, tmp_alias,
154 &dir->i_dentry,d_alias) {
155 if (!list_empty(&dentry->d_subdirs)) {
156 struct dentry *child;
157 list_for_each_entry_safe(child, tmp_subdir,
160 /* XXX Print some debug here? */
162 /* Negative dentry. If we were
163 dropping dcache lock, go
164 throught the list again */
165 if (ll_drop_dentry(child))
170 spin_unlock(&dcache_lock);
171 spin_unlock(&ll_lookup_lock);
175 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
176 void *data, int flag)
179 struct lustre_handle lockh;
183 case LDLM_CB_BLOCKING:
184 ldlm_lock2handle(lock, &lockh);
185 rc = ldlm_cli_cancel(&lockh);
187 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
191 case LDLM_CB_CANCELING: {
192 struct inode *inode = ll_inode_from_lock(lock);
193 __u64 bits = lock->l_policy_data.l_inodebits.bits;
196 /* Invalidate all dentries associated with this inode */
200 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
201 if ((bits & MDS_INODELOCK_LOOKUP) &&
202 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
203 bits &= ~MDS_INODELOCK_LOOKUP;
204 if ((bits & MDS_INODELOCK_UPDATE) &&
205 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
206 bits &= ~MDS_INODELOCK_UPDATE;
207 if ((bits & MDS_INODELOCK_OPEN) &&
208 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
209 bits &= ~MDS_INODELOCK_OPEN;
211 fid = ll_inode2fid(inode);
212 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
213 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
214 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
215 LDLM_ERROR(lock, "data mismatch with object "
216 DFID" (%p)", PFID(fid), inode);
219 if (bits & MDS_INODELOCK_OPEN) {
221 switch (lock->l_req_mode) {
232 CERROR("Unexpected lock mode for OPEN lock "
233 "%d, inode %ld\n", lock->l_req_mode,
236 ll_md_real_close(inode, flags);
239 if (bits & MDS_INODELOCK_UPDATE)
240 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
242 if (S_ISDIR(inode->i_mode) &&
243 (bits & MDS_INODELOCK_UPDATE)) {
244 CDEBUG(D_INODE, "invalidating inode %lu\n",
246 truncate_inode_pages(inode->i_mapping, 0);
247 ll_drop_negative_dentry(inode);
250 if (inode->i_sb->s_root &&
251 inode != inode->i_sb->s_root->d_inode &&
252 (bits & MDS_INODELOCK_LOOKUP))
253 ll_unhash_aliases(inode);
264 __u32 ll_i2suppgid(struct inode *i)
266 if (in_group_p(i->i_gid))
267 return (__u32)i->i_gid;
272 /* Pack the required supplementary groups into the supplied groups array.
273 * If we don't need to use the groups from the target inode(s) then we
274 * instead pack one or more groups from the user's supplementary group
275 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
276 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
283 LASSERT(suppgids != NULL);
285 suppgids[0] = ll_i2suppgid(i1);
288 suppgids[1] = ll_i2suppgid(i2);
293 for (i = 0; i < current_ngroups; i++) {
294 if (suppgids[0] == -1) {
295 if (current_groups[i] != suppgids[1])
296 suppgids[0] = current_groups[i];
299 if (suppgids[1] == -1) {
300 if (current_groups[i] != suppgids[0])
301 suppgids[1] = current_groups[i];
309 static void ll_d_add(struct dentry *de, struct inode *inode)
311 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
313 if (!list_empty(&de->d_alias)) {
314 spin_unlock(&dcache_lock);
315 CERROR("dentry %.*s %p alias next %p, prev %p\n",
316 de->d_name.len, de->d_name.name, de,
317 de->d_alias.next, de->d_alias.prev);
321 list_add(&de->d_alias, &inode->i_dentry);
325 if (!d_unhashed(de)) {
326 spin_unlock(&dcache_lock);
327 CERROR("dentry %.*s %p hash next %p\n",
328 de->d_name.len, de->d_name.name, de, de->d_hash.next);
331 d_rehash_cond(de, 0);
334 /* Search "inode"'s alias list for a dentry that has the same name and parent
335 * as de. If found, return it. If not found, return de.
336 * Lustre can't use d_add_unique because don't unhash aliases for directory
337 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
338 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
340 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
342 struct list_head *tmp;
343 struct dentry *dentry;
344 struct dentry *last_discon = NULL;
346 spin_lock(&ll_lookup_lock);
347 spin_lock(&dcache_lock);
348 list_for_each(tmp, &inode->i_dentry) {
349 dentry = list_entry(tmp, struct dentry, d_alias);
351 /* We are called here with 'de' already on the aliases list. */
352 if (unlikely(dentry == de)) {
357 if (dentry->d_flags & DCACHE_DISCONNECTED) {
358 LASSERT(last_discon == NULL);
359 last_discon = dentry;
363 if (dentry->d_parent != de->d_parent)
366 if (dentry->d_name.hash != de->d_name.hash)
369 if (dentry->d_name.len != de->d_name.len)
372 if (memcmp(dentry->d_name.name, de->d_name.name,
373 de->d_name.len) != 0)
379 #ifdef DCACHE_LUSTRE_INVALID
380 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
382 unlock_dentry(dentry);
383 ll_dops_init(dentry, 0);
384 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
385 spin_unlock(&dcache_lock);
386 spin_unlock(&ll_lookup_lock);
388 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
389 "refc %d\n", de->d_name.len, de->d_name.name, de,
390 de->d_parent, de->d_inode, atomic_read(&de->d_count));
395 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
396 "refc %d\n", last_discon, last_discon->d_inode,
397 atomic_read(&last_discon->d_count));
398 dget_locked(last_discon);
399 spin_unlock(&dcache_lock);
400 spin_unlock(&ll_lookup_lock);
401 ll_dops_init(last_discon, 1);
403 d_move(last_discon, de);
410 spin_unlock(&dcache_lock);
411 spin_unlock(&ll_lookup_lock);
416 int ll_lookup_it_finish(struct ptlrpc_request *request,
417 struct lookup_intent *it, void *data)
419 struct it_cb_data *icbd = data;
420 struct dentry **de = icbd->icbd_childp;
421 struct inode *parent = icbd->icbd_parent;
422 struct ll_sb_info *sbi = ll_i2sbi(parent);
423 struct inode *inode = NULL;
427 /* NB 1 request reference will be taken away by ll_intent_lock()
429 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
430 struct dentry *save = *de;
432 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
436 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
437 inode, inode->i_ino, inode->i_generation);
438 md_set_lock_data(sbi->ll_md_exp,
439 &it->d.lustre.it_lock_handle, inode);
441 /* We used to query real size from OSTs here, but actually
442 this is not needed. For stat() calls size would be updated
443 from subsequent do_revalidate()->ll_inode_revalidate_it() in
445 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
446 Everybody else who needs correct file size would call
447 cl_glimpse_size or some equivalent themselves anyway.
448 Also see bug 7198. */
450 ll_dops_init(*de, 1);
451 *de = ll_find_alias(inode, *de);
453 struct ll_dentry_data *lld = ll_d2d(*de);
455 /* just make sure the ll_dentry_data is ready */
456 if (unlikely(lld == NULL)) {
459 if (likely(lld != NULL))
460 lld->lld_sa_generation = 0;
464 ll_dops_init(*de, 1);
465 /* Check that parent has UPDATE lock. If there is none, we
466 cannot afford to hash this dentry (done by ll_d_add) as it
467 might get picked up later when UPDATE lock will appear */
468 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
469 spin_lock(&dcache_lock);
471 spin_unlock(&dcache_lock);
473 (*de)->d_inode = NULL;
474 /* We do not want to hash the dentry if don`t have a
475 * lock, but if this dentry is later used in d_move,
476 * we'd hit uninitialised list head d_hash, so we just
477 * do this to init d_hash field but leave dentry
478 * unhashed. (bug 10796). */
487 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
488 struct lookup_intent *it, int lookup_flags)
490 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
491 struct dentry *save = dentry, *retval;
492 struct ptlrpc_request *req = NULL;
493 struct md_op_data *op_data;
494 struct it_cb_data icbd;
499 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
500 RETURN(ERR_PTR(-ENAMETOOLONG));
502 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
503 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
504 parent->i_generation, parent, LL_IT2STR(it));
506 if (d_mountpoint(dentry))
507 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
509 ll_frob_intent(&it, &lookup_it);
511 /* As do_lookup is called before follow_mount, root dentry may be left
512 * not valid, revalidate it here. */
513 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
514 (it->it_op & (IT_OPEN | IT_CREAT))) {
515 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
520 if (it->it_op == IT_GETATTR) {
521 first = ll_statahead_enter(parent, &dentry, 1);
523 ll_statahead_exit(dentry, first);
525 RETURN(retval = dentry);
529 icbd.icbd_childp = &dentry;
530 icbd.icbd_parent = parent;
532 if (it->it_op & IT_CREAT ||
533 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
534 opc = LUSTRE_OPC_CREATE;
536 opc = LUSTRE_OPC_ANY;
538 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
539 dentry->d_name.len, lookup_flags, opc,
542 RETURN((void *)op_data);
544 it->it_create_mode &= ~current->fs->umask;
546 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
547 lookup_flags, &req, ll_md_blocking_ast, 0);
548 ll_finish_md_op_data(op_data);
550 GOTO(out, retval = ERR_PTR(rc));
552 rc = ll_lookup_it_finish(req, it, &icbd);
554 ll_intent_release(it);
555 GOTO(out, retval = ERR_PTR(rc));
558 if (first == -EEXIST)
559 ll_statahead_mark(dentry);
561 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
562 !S_ISREG(dentry->d_inode->i_mode) &&
563 !S_ISDIR(dentry->d_inode->i_mode)) {
564 ll_release_openhandle(dentry, it);
566 ll_lookup_finish_locks(it, dentry);
569 GOTO(out, retval = NULL);
571 GOTO(out, retval = dentry);
574 ptlrpc_req_finished(req);
578 #ifdef HAVE_VFS_INTENT_PATCHES
579 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
580 struct nameidata *nd)
585 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
586 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
588 de = ll_lookup_it(parent, dentry, NULL, 0);
593 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
596 struct lookup_intent *it;
598 OBD_ALLOC(it, sizeof(*it));
600 return ERR_PTR(-ENOMEM);
602 if (lookup_flags & LOOKUP_OPEN) {
604 if (lookup_flags & LOOKUP_CREATE)
605 it->it_op |= IT_CREAT;
606 it->it_create_mode = oit->create_mode;
607 it->it_flags = oit->flags;
609 it->it_op = IT_GETATTR;
612 #ifndef HAVE_FILE_IN_STRUCT_INTENT
613 /* Since there is no way to pass our intent to ll_file_open,
614 * just check the file is there. Actual open will be done
616 if (it->it_op & IT_OPEN)
617 it->it_op = IT_LOOKUP;
623 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
624 struct nameidata *nd)
629 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
630 struct lookup_intent *it;
632 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
633 /* Did we came here from failed revalidate just to propagate
635 if (nd->flags & LOOKUP_OPEN)
636 if (IS_ERR(nd->intent.open.file))
637 RETURN((struct dentry *)nd->intent.open.file);
640 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
641 it = ll_d2d(dentry)->lld_it;
642 ll_d2d(dentry)->lld_it = NULL;
644 it = ll_convert_intent(&nd->intent.open, nd->flags);
646 RETURN((struct dentry *)it);
649 de = ll_lookup_it(parent, dentry, it, nd->flags);
652 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
653 if (dentry->d_inode &&
654 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
655 #ifdef HAVE_FILE_IN_STRUCT_INTENT
656 if (S_ISFIFO(dentry->d_inode->i_mode)) {
657 // We cannot call open here as it would
660 (struct ptlrpc_request *)
661 it->d.lustre.it_data);
663 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
664 /* 2.6.1[456] have a bug in open_namei() that forgets to check
665 * nd->intent.open.file for error, so we need to return it as lookup's result
668 nd->intent.open.file->private_data = it;
669 filp =lookup_instantiate_filp(nd,dentry,
674 de = (struct dentry *) filp;
677 nd->intent.open.file->private_data = it;
678 (void)lookup_instantiate_filp(nd,dentry,
683 #else /* HAVE_FILE_IN_STRUCT_INTENT */
684 /* Release open handle as we have no way to
685 * pass it to ll_file_open */
686 ll_release_openhandle(dentry, it);
687 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
688 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
689 // XXX This can only reliably work on assumption
690 // that there are NO hashed negative dentries.
691 ll_d2d(dentry)->lld_it = it;
692 it = NULL; /* Will be freed in ll_create_nd */
693 /* We absolutely depend on ll_create_nd to be
694 * called to not leak this intent and possible
695 * data attached to it */
700 ll_intent_release(it);
701 OBD_FREE(it, sizeof(*it));
704 de = ll_lookup_it(parent, dentry, NULL, 0);
711 /* We depend on "mode" being set with the proper file type/umask by now */
712 static struct inode *ll_create_node(struct inode *dir, const char *name,
713 int namelen, const void *data, int datalen,
714 int mode, __u64 extra,
715 struct lookup_intent *it)
717 struct inode *inode = NULL;
718 struct ptlrpc_request *request = NULL;
719 struct ll_sb_info *sbi = ll_i2sbi(dir);
723 LASSERT(it && it->d.lustre.it_disposition);
725 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
726 request = it->d.lustre.it_data;
727 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
728 rc = ll_prep_inode(&inode, request, dir->i_sb);
730 GOTO(out, inode = ERR_PTR(rc));
732 LASSERT(list_empty(&inode->i_dentry));
734 /* We asked for a lock on the directory, but were granted a
735 * lock on the inode. Since we finally have an inode pointer,
736 * stuff it in the lock. */
737 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
738 inode, inode->i_ino, inode->i_generation);
739 md_set_lock_data(sbi->ll_md_exp,
740 &it->d.lustre.it_lock_handle, inode);
743 ptlrpc_req_finished(request);
748 * By the time this is called, we already have created the directory cache
749 * entry for the new file, but it is so far negative - it has no inode.
751 * We defer creating the OBD object(s) until open, to keep the intent and
752 * non-intent code paths similar, and also because we do not have the MDS
753 * inode number before calling ll_create_node() (which is needed for LOV),
754 * so we would need to do yet another RPC to the MDS to store the LOV EA
755 * data on the MDS. If needed, we would pass the PACKED lmm as data and
756 * lmm_size in datalen (the MDS still has code which will handle that).
758 * If the create succeeds, we fill in the inode information
759 * with d_instantiate().
761 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
762 struct lookup_intent *it)
768 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
769 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
770 dir->i_generation, dir, LL_IT2STR(it));
772 rc = it_open_error(DISP_OPEN_CREATE, it);
776 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
777 NULL, 0, mode, 0, it);
779 RETURN(PTR_ERR(inode));
782 d_instantiate(dentry, inode);
783 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
784 * but some callers, e.g. do_coredump, expect dentry to be hashed after
785 * successful create. Hash it here. */
786 spin_lock(&dcache_lock);
787 if (d_unhashed(dentry))
788 d_rehash_cond(dentry, 0);
789 spin_unlock(&dcache_lock);
793 static void ll_update_times(struct ptlrpc_request *request,
796 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
800 /* mtime is always updated with ctime, but can be set in past.
801 As write and utime(2) may happen within 1 second, and utime's
802 mtime has a priority over write's one, so take mtime from mds
803 for the same ctimes. */
804 if (body->valid & OBD_MD_FLCTIME &&
805 body->ctime >= LTIME_S(inode->i_ctime)) {
806 LTIME_S(inode->i_ctime) = body->ctime;
808 if (body->valid & OBD_MD_FLMTIME) {
809 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
810 "to "LPU64"\n", inode->i_ino,
811 LTIME_S(inode->i_mtime), body->mtime);
812 LTIME_S(inode->i_mtime) = body->mtime;
817 static int ll_new_node(struct inode *dir, struct qstr *name,
818 const char *tgt, int mode, int rdev,
819 struct dentry *dchild, __u32 opc)
821 struct ptlrpc_request *request = NULL;
822 struct md_op_data *op_data;
823 struct inode *inode = NULL;
824 struct ll_sb_info *sbi = ll_i2sbi(dir);
829 if (unlikely(tgt != NULL))
830 tgt_len = strlen(tgt) + 1;
832 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
833 name->len, 0, opc, NULL);
835 GOTO(err_exit, err = PTR_ERR(op_data));
837 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
838 current->fsuid, current->fsgid,
839 cfs_curproc_cap_pack(), rdev, &request);
840 ll_finish_md_op_data(op_data);
844 ll_update_times(request, dir);
847 err = ll_prep_inode(&inode, request, dchild->d_sb);
852 d_instantiate(dchild, inode);
856 ptlrpc_req_finished(request);
861 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
862 unsigned rdev, struct dentry *dchild)
867 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
868 name->len, name->name, dir->i_ino, dir->i_generation, dir,
871 mode &= ~current->fs->umask;
873 switch (mode & S_IFMT) {
875 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
881 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
893 #ifndef HAVE_VFS_INTENT_PATCHES
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 #ifdef HAVE_FILE_IN_STRUCT_INTENT
913 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
914 nd->intent.open.file->private_data = it;
915 lookup_instantiate_filp(nd, dentry, NULL);
918 ll_release_openhandle(dentry,it);
922 ll_intent_release(it);
923 OBD_FREE(it, sizeof(*it));
928 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
929 int mode, struct nameidata *nd)
931 if (!nd || !nd->intent.d.lustre.it_disposition)
932 /* No saved request? Just mknod the file */
933 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
935 return ll_create_it(dir, dentry, mode, &nd->intent);
939 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
940 const char *tgt, struct dentry *dchild)
945 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
946 name->len, name->name, dir->i_ino, dir->i_generation,
949 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
950 0, dchild, LUSTRE_OPC_SYMLINK);
954 static int ll_link_generic(struct inode *src, struct inode *dir,
955 struct qstr *name, struct dentry *dchild)
957 struct ll_sb_info *sbi = ll_i2sbi(dir);
958 struct ptlrpc_request *request = NULL;
959 struct md_op_data *op_data;
964 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
965 src->i_ino, src->i_generation, src, dir->i_ino,
966 dir->i_generation, dir, name->len, name->name);
968 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
969 0, LUSTRE_OPC_ANY, NULL);
971 RETURN(PTR_ERR(op_data));
973 err = md_link(sbi->ll_md_exp, op_data, &request);
974 ll_finish_md_op_data(op_data);
980 ll_update_times(request, dir);
983 ptlrpc_req_finished(request);
987 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
988 int mode, struct dentry *dchild)
994 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
995 name->len, name->name, dir->i_ino, dir->i_generation, dir);
997 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
998 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1003 /* Try to find the child dentry by its name.
1004 If found, put the result fid into @fid. */
1005 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1008 struct dentry *parent, *child;
1010 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1011 child = d_lookup(parent, name);
1014 *fid = *ll_inode2fid(child->d_inode);
1019 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1020 struct dentry *dchild, struct qstr *name)
1022 struct ptlrpc_request *request = NULL;
1023 struct md_op_data *op_data;
1027 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1028 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1030 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1033 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1034 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1035 if (IS_ERR(op_data))
1036 RETURN(PTR_ERR(op_data));
1038 ll_get_child_fid(dir, name, &op_data->op_fid3);
1039 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1040 ll_finish_md_op_data(op_data);
1042 ll_update_times(request, dir);
1043 ptlrpc_req_finished(request);
1047 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1049 struct mdt_body *body;
1050 struct lov_mds_md *eadata;
1051 struct lov_stripe_md *lsm = NULL;
1052 struct obd_trans_info oti = { 0 };
1054 struct obd_capa *oc = NULL;
1058 /* req is swabbed so this is safe */
1059 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1060 if (!(body->valid & OBD_MD_FLEASIZE))
1063 if (body->eadatasize == 0) {
1064 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1065 GOTO(out, rc = -EPROTO);
1068 /* The MDS sent back the EA because we unlinked the last reference
1069 * to this file. Use this EA to unlink the objects on the OST.
1070 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1071 * check it is complete and sensible. */
1072 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1074 LASSERT(eadata != NULL);
1076 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1078 CERROR("obd_unpackmd: %d\n", rc);
1081 LASSERT(rc >= sizeof(*lsm));
1083 rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1085 GOTO(out_free_memmd, rc);
1089 GOTO(out_free_memmd, rc = -ENOMEM);
1091 oa->o_id = lsm->lsm_object_id;
1092 oa->o_gr = lsm->lsm_object_gr;
1093 oa->o_mode = body->mode & S_IFMT;
1094 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1096 if (body->valid & OBD_MD_FLCOOKIE) {
1097 oa->o_valid |= OBD_MD_FLCOOKIE;
1098 oti.oti_logcookies =
1099 req_capsule_server_sized_get(&request->rq_pill,
1101 sizeof(struct llog_cookie) *
1102 lsm->lsm_stripe_count);
1103 if (oti.oti_logcookies == NULL) {
1104 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1105 body->valid &= ~OBD_MD_FLCOOKIE;
1109 if (body->valid & OBD_MD_FLOSSCAPA) {
1110 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1112 GOTO(out_free_memmd, rc);
1115 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1119 CERROR("obd destroy objid "LPX64" error %d\n",
1120 lsm->lsm_object_id, rc);
1122 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1127 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1128 struct dentry *dchild, struct qstr *name)
1130 struct ptlrpc_request *request = NULL;
1131 struct md_op_data *op_data;
1134 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1135 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1138 * XXX: unlink bind mountpoint maybe call to here,
1139 * just check it as vfs_unlink does.
1141 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1144 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1145 name->len, 0, LUSTRE_OPC_ANY, NULL);
1146 if (IS_ERR(op_data))
1147 RETURN(PTR_ERR(op_data));
1149 ll_get_child_fid(dir, name, &op_data->op_fid3);
1150 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1151 ll_finish_md_op_data(op_data);
1155 ll_update_times(request, dir);
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 err = ll_objects_destroy(request, src);
1199 ptlrpc_req_finished(request);
1204 #ifdef HAVE_VFS_INTENT_PATCHES
1205 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1207 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1209 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1211 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1213 tgtnd->dentry->d_inode, tgtnd->dentry,
1214 NULL, &tgtnd->last);
1216 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1218 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1219 &tgtnd->last, NULL);
1221 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1223 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1225 static int ll_rmdir_raw(struct nameidata *nd)
1227 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1230 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1232 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1234 static int ll_unlink_raw(struct nameidata *nd)
1236 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1241 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1244 return ll_mknod_generic(dir, &dchild->d_name, mode,
1245 old_encode_dev(rdev), dchild);
1248 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1250 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1252 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1254 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1256 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1258 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1260 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1261 const char *oldname)
1263 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1265 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1266 struct dentry *new_dentry)
1268 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1271 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1272 struct inode *new_dir, struct dentry *new_dentry)
1275 err = ll_rename_generic(old_dir, NULL,
1276 old_dentry, &old_dentry->d_name,
1277 new_dir, NULL, new_dentry,
1278 &new_dentry->d_name);
1280 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1281 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1283 d_move(old_dentry, new_dentry);
1288 struct inode_operations ll_dir_inode_operations = {
1289 #ifdef HAVE_VFS_INTENT_PATCHES
1290 .link_raw = ll_link_raw,
1291 .unlink_raw = ll_unlink_raw,
1292 .symlink_raw = ll_symlink_raw,
1293 .mkdir_raw = ll_mkdir_raw,
1294 .rmdir_raw = ll_rmdir_raw,
1295 .mknod_raw = ll_mknod_raw,
1296 .rename_raw = ll_rename_raw,
1297 .setattr = ll_setattr,
1298 .setattr_raw = ll_setattr_raw,
1301 .lookup = ll_lookup_nd,
1302 .create = ll_create_nd,
1303 /* We need all these non-raw things for NFSD, to not patch it. */
1304 .unlink = ll_unlink,
1307 .symlink = ll_symlink,
1309 .rename = ll_rename,
1310 .setattr = ll_setattr,
1311 .getattr = ll_getattr,
1312 .permission = ll_inode_permission,
1313 .setxattr = ll_setxattr,
1314 .getxattr = ll_getxattr,
1315 .listxattr = ll_listxattr,
1316 .removexattr = ll_removexattr,
1319 struct inode_operations ll_special_inode_operations = {
1320 #ifdef HAVE_VFS_INTENT_PATCHES
1321 .setattr_raw = ll_setattr_raw,
1323 .setattr = ll_setattr,
1324 .getattr = ll_getattr,
1325 .permission = ll_inode_permission,
1326 .setxattr = ll_setxattr,
1327 .getxattr = ll_getxattr,
1328 .listxattr = ll_listxattr,
1329 .removexattr = ll_removexattr,