4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 #include <linux/sched.h>
40 #include <linux/quotaops.h>
41 #include <linux/highmem.h>
42 #include <linux/pagemap.h>
43 #include <linux/security.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 "llite_internal.h"
55 * Check if we have something mounted at the named dchild.
56 * In such a case there would always be dentry present.
58 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
63 if (unlikely(dchild)) {
64 mounted = d_mountpoint(dchild);
66 dchild = d_lookup(dparent, name);
68 mounted = d_mountpoint(dchild);
75 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
79 ldlm_lock_decref(lockh, mode);
85 /* called from iget5_locked->find_inode() under inode_lock spinlock */
86 static int ll_test_inode(struct inode *inode, void *opaque)
88 struct ll_inode_info *lli = ll_i2info(inode);
89 struct lustre_md *md = opaque;
91 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
92 CERROR("MDS body missing FID\n");
96 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
102 static int ll_set_inode(struct inode *inode, void *opaque)
104 struct ll_inode_info *lli = ll_i2info(inode);
105 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
107 if (unlikely(!(body->valid & OBD_MD_FLID))) {
108 CERROR("MDS body missing FID\n");
112 lli->lli_fid = body->fid1;
113 if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
114 CERROR("Can not initialize inode "DFID" without object type: "
115 "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
119 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
120 if (unlikely(inode->i_mode == 0)) {
121 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
132 * Get an inode by inode number (already instantiated by the intent lookup).
133 * Returns inode or NULL
135 struct inode *ll_iget(struct super_block *sb, ino_t hash,
136 struct lustre_md *md)
142 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
145 if (inode->i_state & I_NEW) {
148 ll_read_inode2(inode, md);
149 if (S_ISREG(inode->i_mode) &&
150 ll_i2info(inode)->lli_clob == NULL)
151 rc = cl_file_inode_init(inode, md);
154 make_bad_inode(inode);
155 unlock_new_inode(inode);
159 unlock_new_inode(inode);
160 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
161 ll_update_inode(inode, md);
162 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
163 inode, PFID(&md->body->fid1));
168 static void ll_invalidate_negative_children(struct inode *dir)
170 struct dentry *dentry, *tmp_subdir;
173 list_for_each_entry(dentry, &dir->i_dentry, d_alias) {
174 spin_lock(&dentry->d_lock);
175 if (!list_empty(&dentry->d_subdirs)) {
176 struct dentry *child;
178 list_for_each_entry_safe(child, tmp_subdir,
181 if (child->d_inode == NULL)
182 d_lustre_invalidate(child);
185 spin_unlock(&dentry->d_lock);
187 ll_unlock_dcache(dir);
190 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
191 void *data, int flag)
194 struct lustre_handle lockh;
198 case LDLM_CB_BLOCKING:
199 ldlm_lock2handle(lock, &lockh);
200 rc = ldlm_cli_cancel(&lockh);
202 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
206 case LDLM_CB_CANCELING: {
207 struct inode *inode = ll_inode_from_lock(lock);
208 struct ll_inode_info *lli;
209 __u64 bits = lock->l_policy_data.l_inodebits.bits;
211 ldlm_mode_t mode = lock->l_req_mode;
213 /* Invalidate all dentries associated with this inode */
217 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
218 /* For OPEN locks we differentiate between lock modes
219 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
220 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
221 MDS_INODELOCK_LAYOUT))
222 ll_have_md_lock(inode, &bits, LCK_MINMODE);
224 if (bits & MDS_INODELOCK_OPEN)
225 ll_have_md_lock(inode, &bits, mode);
227 fid = ll_inode2fid(inode);
228 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
229 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
230 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
231 LDLM_ERROR(lock, "data mismatch with object "
232 DFID" (%p)", PFID(fid), inode);
235 if (bits & MDS_INODELOCK_OPEN) {
237 switch (lock->l_req_mode) {
248 CERROR("Unexpected lock mode for OPEN lock "
249 "%d, inode %ld\n", lock->l_req_mode,
252 ll_md_real_close(inode, flags);
255 lli = ll_i2info(inode);
256 if (bits & MDS_INODELOCK_LAYOUT) {
257 struct cl_object_conf conf = { .coc_inode = inode,
258 .coc_invalidate = true };
259 rc = ll_layout_conf(inode, &conf);
261 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
264 if (bits & MDS_INODELOCK_UPDATE)
265 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
267 if (S_ISDIR(inode->i_mode) &&
268 (bits & MDS_INODELOCK_UPDATE)) {
269 CDEBUG(D_INODE, "invalidating inode %lu\n",
271 truncate_inode_pages(inode->i_mapping, 0);
272 ll_invalidate_negative_children(inode);
275 if (inode->i_sb->s_root &&
276 inode != inode->i_sb->s_root->d_inode &&
277 (bits & MDS_INODELOCK_LOOKUP))
278 ll_invalidate_aliases(inode);
289 __u32 ll_i2suppgid(struct inode *i)
291 if (cfs_curproc_is_in_groups(i->i_gid))
292 return (__u32)i->i_gid;
297 /* Pack the required supplementary groups into the supplied groups array.
298 * If we don't need to use the groups from the target inode(s) then we
299 * instead pack one or more groups from the user's supplementary group
300 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
301 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
308 LASSERT(suppgids != NULL);
310 suppgids[0] = ll_i2suppgid(i1);
313 suppgids[1] = ll_i2suppgid(i2);
318 for (i = 0; i < current_ngroups; i++) {
319 if (suppgids[0] == -1) {
320 if (current_groups[i] != suppgids[1])
321 suppgids[0] = current_groups[i];
324 if (suppgids[1] == -1) {
325 if (current_groups[i] != suppgids[0])
326 suppgids[1] = current_groups[i];
335 * try to reuse three types of dentry:
336 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
337 * by concurrent .revalidate).
338 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
339 * be cleared by others calling d_lustre_revalidate).
340 * 3. DISCONNECTED alias.
342 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
344 struct dentry *alias, *discon_alias, *invalid_alias;
346 if (list_empty(&inode->i_dentry))
349 discon_alias = invalid_alias = NULL;
351 ll_lock_dcache(inode);
352 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
353 LASSERT(alias != dentry);
355 spin_lock(&alias->d_lock);
356 if (alias->d_flags & DCACHE_DISCONNECTED)
357 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
358 discon_alias = alias;
359 else if (alias->d_parent == dentry->d_parent &&
360 alias->d_name.hash == dentry->d_name.hash &&
361 alias->d_name.len == dentry->d_name.len &&
362 memcmp(alias->d_name.name, dentry->d_name.name,
363 dentry->d_name.len) == 0)
364 invalid_alias = alias;
365 spin_unlock(&alias->d_lock);
370 alias = invalid_alias ?: discon_alias ?: NULL;
372 spin_lock(&alias->d_lock);
374 spin_unlock(&alias->d_lock);
376 ll_unlock_dcache(inode);
382 * Similar to d_splice_alias(), but lustre treats DCACHE_LUSTRE_INVALID alias
383 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
385 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
390 new = ll_find_alias(inode, de);
392 ll_dops_init(new, 1, 1);
396 "Reuse dentry %p inode %p refc %d flags %#x\n",
397 new, new->d_inode, d_refcount(new), new->d_flags);
401 __d_lustre_invalidate(de);
402 ll_dops_init(de, 1, 1);
404 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
405 de, de->d_inode, d_refcount(de), de->d_flags);
409 int ll_lookup_it_finish(struct ptlrpc_request *request,
410 struct lookup_intent *it, void *data)
412 struct it_cb_data *icbd = data;
413 struct dentry **de = icbd->icbd_childp;
414 struct inode *parent = icbd->icbd_parent;
415 struct inode *inode = NULL;
420 /* NB 1 request reference will be taken away by ll_intent_lock()
422 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
423 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
427 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
429 /* We used to query real size from OSTs here, but actually
430 this is not needed. For stat() calls size would be updated
431 from subsequent do_revalidate()->ll_inode_revalidate_it() in
433 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
434 Everybody else who needs correct file size would call
435 ll_glimpse_size or some equivalent themselves anyway.
436 Also see bug 7198. */
439 *de = ll_splice_alias(inode, *de);
441 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
442 /* we have lookup look - unhide dentry */
443 if (bits & MDS_INODELOCK_LOOKUP)
444 d_lustre_revalidate(*de);
446 /* Check that parent has UPDATE lock. */
447 struct lookup_intent parent_it = {
449 .d.lustre.it_lock_handle = 0 };
451 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
452 &ll_i2info(parent)->lli_fid, NULL)) {
453 d_lustre_revalidate(*de);
454 ll_intent_release(&parent_it);
461 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
462 struct lookup_intent *it, int lookup_flags)
464 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
465 struct dentry *save = dentry, *retval;
466 struct ptlrpc_request *req = NULL;
467 struct md_op_data *op_data;
468 struct it_cb_data icbd;
473 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
474 RETURN(ERR_PTR(-ENAMETOOLONG));
476 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
477 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
478 parent->i_generation, parent, LL_IT2STR(it));
480 if (d_mountpoint(dentry))
481 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
483 ll_frob_intent(&it, &lookup_it);
485 /* As do_lookup is called before follow_mount, root dentry may be left
486 * not valid, revalidate it here. */
487 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
488 (it->it_op & (IT_OPEN | IT_CREAT))) {
489 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
490 MDS_INODELOCK_LOOKUP);
495 if (it->it_op == IT_GETATTR) {
496 rc = ll_statahead_enter(parent, &dentry, 0);
499 GOTO(out, retval = NULL);
500 GOTO(out, retval = dentry);
504 icbd.icbd_childp = &dentry;
505 icbd.icbd_parent = parent;
507 if (it->it_op & IT_CREAT ||
508 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
509 opc = LUSTRE_OPC_CREATE;
511 opc = LUSTRE_OPC_ANY;
513 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
514 dentry->d_name.len, lookup_flags, opc,
517 RETURN((void *)op_data);
519 it->it_create_mode &= ~cfs_curproc_umask();
521 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
522 lookup_flags, &req, ll_md_blocking_ast, 0);
523 ll_finish_md_op_data(op_data);
525 GOTO(out, retval = ERR_PTR(rc));
527 rc = ll_lookup_it_finish(req, it, &icbd);
529 ll_intent_release(it);
530 GOTO(out, retval = ERR_PTR(rc));
533 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
534 !S_ISREG(dentry->d_inode->i_mode) &&
535 !S_ISDIR(dentry->d_inode->i_mode)) {
536 ll_release_openhandle(dentry, it);
538 ll_lookup_finish_locks(it, dentry);
541 GOTO(out, retval = NULL);
543 GOTO(out, retval = dentry);
546 ptlrpc_req_finished(req);
547 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
548 ll_statahead_mark(parent, dentry);
552 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
555 struct lookup_intent *it;
557 OBD_ALLOC(it, sizeof(*it));
559 return ERR_PTR(-ENOMEM);
561 if (lookup_flags & LOOKUP_OPEN) {
563 if (lookup_flags & LOOKUP_CREATE)
564 it->it_op |= IT_CREAT;
565 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
566 it->it_flags = oit->flags;
568 it->it_op = IT_GETATTR;
574 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
575 struct nameidata *nd)
580 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
581 struct lookup_intent *it;
583 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
584 it = ll_d2d(dentry)->lld_it;
585 ll_d2d(dentry)->lld_it = NULL;
587 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
588 /* We are sure this is new dentry, so we need to create
589 our private data and set the dentry ops */
590 ll_dops_init(dentry, 1, 1);
591 __d_lustre_invalidate(dentry);
595 it = ll_convert_intent(&nd->intent.open, nd->flags);
597 RETURN((struct dentry *)it);
600 de = ll_lookup_it(parent, dentry, it, nd->flags);
603 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
604 if (dentry->d_inode &&
605 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
606 if (S_ISFIFO(dentry->d_inode->i_mode)) {
607 // We cannot call open here as it would
610 (struct ptlrpc_request *)
611 it->d.lustre.it_data);
615 nd->intent.open.file->private_data = it;
616 filp = lookup_instantiate_filp(nd,
622 de = (struct dentry *)filp;
625 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
626 // XXX This can only reliably work on assumption
627 // that there are NO hashed negative dentries.
628 ll_d2d(dentry)->lld_it = it;
629 it = NULL; /* Will be freed in ll_create_nd */
630 /* We absolutely depend on ll_create_nd to be
631 * called to not leak this intent and possible
632 * data attached to it */
637 ll_intent_release(it);
638 OBD_FREE(it, sizeof(*it));
641 de = ll_lookup_it(parent, dentry, NULL, 0);
647 /* We depend on "mode" being set with the proper file type/umask by now */
648 static struct inode *ll_create_node(struct inode *dir, const char *name,
649 int namelen, const void *data, int datalen,
650 int mode, __u64 extra,
651 struct lookup_intent *it)
653 struct inode *inode = NULL;
654 struct ptlrpc_request *request = NULL;
655 struct ll_sb_info *sbi = ll_i2sbi(dir);
659 LASSERT(it && it->d.lustre.it_disposition);
661 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
662 request = it->d.lustre.it_data;
663 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
664 rc = ll_prep_inode(&inode, request, dir->i_sb);
666 GOTO(out, inode = ERR_PTR(rc));
668 LASSERT(list_empty(&inode->i_dentry));
670 /* We asked for a lock on the directory, but were granted a
671 * lock on the inode. Since we finally have an inode pointer,
672 * stuff it in the lock. */
673 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
674 inode, inode->i_ino, inode->i_generation);
675 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
678 ptlrpc_req_finished(request);
683 * By the time this is called, we already have created the directory cache
684 * entry for the new file, but it is so far negative - it has no inode.
686 * We defer creating the OBD object(s) until open, to keep the intent and
687 * non-intent code paths similar, and also because we do not have the MDS
688 * inode number before calling ll_create_node() (which is needed for LOV),
689 * so we would need to do yet another RPC to the MDS to store the LOV EA
690 * data on the MDS. If needed, we would pass the PACKED lmm as data and
691 * lmm_size in datalen (the MDS still has code which will handle that).
693 * If the create succeeds, we fill in the inode information
694 * with d_instantiate().
696 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
697 struct lookup_intent *it)
703 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
704 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
705 dir->i_generation, dir, LL_IT2STR(it));
707 rc = it_open_error(DISP_OPEN_CREATE, it);
711 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
712 NULL, 0, mode, 0, it);
714 RETURN(PTR_ERR(inode));
716 d_instantiate(dentry, inode);
720 static void ll_update_times(struct ptlrpc_request *request,
723 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
727 if (body->valid & OBD_MD_FLMTIME &&
728 body->mtime > LTIME_S(inode->i_mtime)) {
729 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
730 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
731 LTIME_S(inode->i_mtime) = body->mtime;
733 if (body->valid & OBD_MD_FLCTIME &&
734 body->ctime > LTIME_S(inode->i_ctime))
735 LTIME_S(inode->i_ctime) = body->ctime;
738 static int ll_new_node(struct inode *dir, struct qstr *name,
739 const char *tgt, int mode, int rdev,
740 struct dentry *dchild, __u32 opc)
742 struct ptlrpc_request *request = NULL;
743 struct md_op_data *op_data;
744 struct inode *inode = NULL;
745 struct ll_sb_info *sbi = ll_i2sbi(dir);
750 if (unlikely(tgt != NULL))
751 tgt_len = strlen(tgt) + 1;
753 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
754 name->len, 0, opc, NULL);
756 GOTO(err_exit, err = PTR_ERR(op_data));
758 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
759 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
760 cfs_curproc_cap_pack(), rdev, &request);
761 ll_finish_md_op_data(op_data);
765 ll_update_times(request, dir);
768 err = ll_prep_inode(&inode, request, dchild->d_sb);
772 d_instantiate(dchild, inode);
776 ptlrpc_req_finished(request);
781 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
782 unsigned rdev, struct dentry *dchild)
787 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
788 name->len, name->name, dir->i_ino, dir->i_generation, dir,
791 mode &= ~cfs_curproc_umask();
793 switch (mode & S_IFMT) {
795 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
801 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
812 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
817 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
818 int mode, struct nameidata *nd)
820 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
824 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
826 ll_d2d(dentry)->lld_it = NULL;
828 /* Was there an error? Propagate it! */
829 if (it->d.lustre.it_status) {
830 rc = it->d.lustre.it_status;
834 rc = ll_create_it(dir, dentry, mode, it);
835 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
838 nd->intent.open.file->private_data = it;
839 filp = lookup_instantiate_filp(nd, dentry, NULL);
845 ll_intent_release(it);
846 OBD_FREE(it, sizeof(*it));
849 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
854 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
855 const char *tgt, struct dentry *dchild)
860 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
861 name->len, name->name, dir->i_ino, dir->i_generation,
864 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
865 0, dchild, LUSTRE_OPC_SYMLINK);
868 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
873 static int ll_link_generic(struct inode *src, struct inode *dir,
874 struct qstr *name, struct dentry *dchild)
876 struct ll_sb_info *sbi = ll_i2sbi(dir);
877 struct ptlrpc_request *request = NULL;
878 struct md_op_data *op_data;
883 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
884 src->i_ino, src->i_generation, src, dir->i_ino,
885 dir->i_generation, dir, name->len, name->name);
887 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
888 0, LUSTRE_OPC_ANY, NULL);
890 RETURN(PTR_ERR(op_data));
892 err = md_link(sbi->ll_md_exp, op_data, &request);
893 ll_finish_md_op_data(op_data);
897 ll_update_times(request, dir);
898 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
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) & ~cfs_curproc_umask()) | S_IFDIR;
916 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
919 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
924 /* Try to find the child dentry by its name.
925 If found, put the result fid into @fid. */
926 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
929 struct dentry *parent, *child;
931 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
932 child = d_lookup(parent, name);
935 *fid = *ll_inode2fid(child->d_inode);
940 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
941 struct dentry *dchild, struct qstr *name)
943 struct ptlrpc_request *request = NULL;
944 struct md_op_data *op_data;
948 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
949 name->len, name->name, dir->i_ino, dir->i_generation, dir);
951 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
954 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
955 S_IFDIR, LUSTRE_OPC_ANY, NULL);
957 RETURN(PTR_ERR(op_data));
959 ll_get_child_fid(dir, name, &op_data->op_fid3);
960 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
961 ll_finish_md_op_data(op_data);
963 ll_update_times(request, dir);
964 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
967 ptlrpc_req_finished(request);
971 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
973 struct mdt_body *body;
974 struct lov_mds_md *eadata;
975 struct lov_stripe_md *lsm = NULL;
976 struct obd_trans_info oti = { 0 };
978 struct obd_capa *oc = NULL;
982 /* req is swabbed so this is safe */
983 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
984 if (!(body->valid & OBD_MD_FLEASIZE))
987 if (body->eadatasize == 0) {
988 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
989 GOTO(out, rc = -EPROTO);
992 /* The MDS sent back the EA because we unlinked the last reference
993 * to this file. Use this EA to unlink the objects on the OST.
994 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
995 * check it is complete and sensible. */
996 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
998 LASSERT(eadata != NULL);
1000 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1002 CERROR("obd_unpackmd: %d\n", rc);
1005 LASSERT(rc >= sizeof(*lsm));
1009 GOTO(out_free_memmd, rc = -ENOMEM);
1011 oa->o_id = lsm->lsm_object_id;
1012 oa->o_seq = lsm->lsm_object_seq;
1013 oa->o_mode = body->mode & S_IFMT;
1014 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1016 if (body->valid & OBD_MD_FLCOOKIE) {
1017 oa->o_valid |= OBD_MD_FLCOOKIE;
1018 oti.oti_logcookies =
1019 req_capsule_server_sized_get(&request->rq_pill,
1021 sizeof(struct llog_cookie) *
1022 lsm->lsm_stripe_count);
1023 if (oti.oti_logcookies == NULL) {
1024 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1025 body->valid &= ~OBD_MD_FLCOOKIE;
1029 if (body->valid & OBD_MD_FLOSSCAPA) {
1030 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1032 GOTO(out_free_memmd, rc);
1035 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1036 ll_i2mdexp(dir), oc);
1040 CERROR("obd destroy objid "LPX64" error %d\n",
1041 lsm->lsm_object_id, rc);
1043 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1048 /* ll_unlink_generic() doesn't update the inode with the new link count.
1049 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1050 * is any lock existing. They will recycle dentries and inodes based upon locks
1052 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1053 struct dentry *dchild, struct qstr *name)
1055 struct ptlrpc_request *request = NULL;
1056 struct md_op_data *op_data;
1059 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1060 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1063 * XXX: unlink bind mountpoint maybe call to here,
1064 * just check it as vfs_unlink does.
1066 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1069 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1070 name->len, 0, LUSTRE_OPC_ANY, NULL);
1071 if (IS_ERR(op_data))
1072 RETURN(PTR_ERR(op_data));
1074 ll_get_child_fid(dir, name, &op_data->op_fid3);
1075 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1076 ll_finish_md_op_data(op_data);
1080 ll_update_times(request, dir);
1081 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1083 rc = ll_objects_destroy(request, dir);
1085 ptlrpc_req_finished(request);
1089 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1090 struct dentry *src_dchild, struct qstr *src_name,
1091 struct inode *tgt, struct dentry *tgt_dparent,
1092 struct dentry *tgt_dchild, struct qstr *tgt_name)
1094 struct ptlrpc_request *request = NULL;
1095 struct ll_sb_info *sbi = ll_i2sbi(src);
1096 struct md_op_data *op_data;
1099 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1100 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1101 src->i_ino, src->i_generation, src, tgt_name->len,
1102 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1104 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1105 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1108 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1109 LUSTRE_OPC_ANY, NULL);
1110 if (IS_ERR(op_data))
1111 RETURN(PTR_ERR(op_data));
1113 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1114 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1115 err = md_rename(sbi->ll_md_exp, op_data,
1116 src_name->name, src_name->len,
1117 tgt_name->name, tgt_name->len, &request);
1118 ll_finish_md_op_data(op_data);
1120 ll_update_times(request, src);
1121 ll_update_times(request, tgt);
1122 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1123 err = ll_objects_destroy(request, src);
1126 ptlrpc_req_finished(request);
1131 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1134 return ll_mknod_generic(dir, &dchild->d_name, mode,
1135 old_encode_dev(rdev), dchild);
1138 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1140 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1142 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1144 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1146 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1148 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1150 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1151 const char *oldname)
1153 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1155 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1156 struct dentry *new_dentry)
1158 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1161 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1162 struct inode *new_dir, struct dentry *new_dentry)
1165 err = ll_rename_generic(old_dir, NULL,
1166 old_dentry, &old_dentry->d_name,
1167 new_dir, NULL, new_dentry,
1168 &new_dentry->d_name);
1170 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1171 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1173 d_move(old_dentry, new_dentry);
1178 struct inode_operations ll_dir_inode_operations = {
1180 .lookup = ll_lookup_nd,
1181 .create = ll_create_nd,
1182 /* We need all these non-raw things for NFSD, to not patch it. */
1183 .unlink = ll_unlink,
1186 .symlink = ll_symlink,
1188 .rename = ll_rename,
1189 .setattr = ll_setattr,
1190 .getattr = ll_getattr,
1191 .permission = ll_inode_permission,
1192 .setxattr = ll_setxattr,
1193 .getxattr = ll_getxattr,
1194 .listxattr = ll_listxattr,
1195 .removexattr = ll_removexattr,
1198 struct inode_operations ll_special_inode_operations = {
1199 .setattr = ll_setattr,
1200 .getattr = ll_getattr,
1201 .permission = ll_inode_permission,
1202 .setxattr = ll_setxattr,
1203 .getxattr = ll_getxattr,
1204 .listxattr = ll_listxattr,
1205 .removexattr = ll_removexattr,