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, Intel Corporation.
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"
54 static int ll_create_it(struct inode *, struct dentry *,
55 int, struct lookup_intent *);
58 * Check if we have something mounted at the named dchild.
59 * In such a case there would always be dentry present.
61 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
66 if (unlikely(dchild)) {
67 mounted = d_mountpoint(dchild);
69 dchild = d_lookup(dparent, name);
71 mounted = d_mountpoint(dchild);
78 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
82 ldlm_lock_decref(lockh, mode);
88 /* called from iget5_locked->find_inode() under inode_lock spinlock */
89 static int ll_test_inode(struct inode *inode, void *opaque)
91 struct ll_inode_info *lli = ll_i2info(inode);
92 struct lustre_md *md = opaque;
94 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
95 CERROR("MDS body missing FID\n");
99 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
105 static int ll_set_inode(struct inode *inode, void *opaque)
107 struct ll_inode_info *lli = ll_i2info(inode);
108 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
110 if (unlikely(!(body->valid & OBD_MD_FLID))) {
111 CERROR("MDS body missing FID\n");
115 lli->lli_fid = body->fid1;
116 if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
117 CERROR("Can not initialize inode "DFID" without object type: "
118 "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
122 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
123 if (unlikely(inode->i_mode == 0)) {
124 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
135 * Get an inode by inode number (already instantiated by the intent lookup).
136 * Returns inode or NULL
138 struct inode *ll_iget(struct super_block *sb, ino_t hash,
139 struct lustre_md *md)
145 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
148 if (inode->i_state & I_NEW) {
151 ll_read_inode2(inode, md);
152 if (S_ISREG(inode->i_mode) &&
153 ll_i2info(inode)->lli_clob == NULL)
154 rc = cl_file_inode_init(inode, md);
157 make_bad_inode(inode);
158 unlock_new_inode(inode);
162 unlock_new_inode(inode);
163 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
164 ll_update_inode(inode, md);
165 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
166 inode, PFID(&md->body->fid1));
171 static void ll_invalidate_negative_children(struct inode *dir)
173 struct dentry *dentry, *tmp_subdir;
174 struct ll_d_hlist_node *p;
177 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
178 spin_lock(&dentry->d_lock);
179 if (!list_empty(&dentry->d_subdirs)) {
180 struct dentry *child;
182 list_for_each_entry_safe(child, tmp_subdir,
185 if (child->d_inode == NULL)
186 d_lustre_invalidate(child);
189 spin_unlock(&dentry->d_lock);
191 ll_unlock_dcache(dir);
194 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
195 void *data, int flag)
198 struct lustre_handle lockh;
202 case LDLM_CB_BLOCKING:
203 ldlm_lock2handle(lock, &lockh);
204 rc = ldlm_cli_cancel(&lockh);
206 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
210 case LDLM_CB_CANCELING: {
211 struct inode *inode = ll_inode_from_lock(lock);
212 struct ll_inode_info *lli;
213 __u64 bits = lock->l_policy_data.l_inodebits.bits;
215 ldlm_mode_t mode = lock->l_req_mode;
217 /* Invalidate all dentries associated with this inode */
221 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
222 /* For OPEN locks we differentiate between lock modes
223 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
224 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
225 MDS_INODELOCK_LAYOUT))
226 ll_have_md_lock(inode, &bits, LCK_MINMODE);
228 if (bits & MDS_INODELOCK_OPEN)
229 ll_have_md_lock(inode, &bits, mode);
231 fid = ll_inode2fid(inode);
232 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
233 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
234 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
235 LDLM_ERROR(lock, "data mismatch with object "
236 DFID" (%p)", PFID(fid), inode);
239 if (bits & MDS_INODELOCK_OPEN) {
241 switch (lock->l_req_mode) {
252 CERROR("Unexpected lock mode for OPEN lock "
253 "%d, inode %ld\n", lock->l_req_mode,
256 ll_md_real_close(inode, flags);
259 lli = ll_i2info(inode);
260 if (bits & MDS_INODELOCK_LAYOUT) {
261 struct cl_object_conf conf = { .coc_inode = inode,
262 .coc_invalidate = true };
263 rc = ll_layout_conf(inode, &conf);
265 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
268 if (bits & MDS_INODELOCK_UPDATE)
269 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
271 if (S_ISDIR(inode->i_mode) &&
272 (bits & MDS_INODELOCK_UPDATE)) {
273 CDEBUG(D_INODE, "invalidating inode %lu\n",
275 truncate_inode_pages(inode->i_mapping, 0);
276 ll_invalidate_negative_children(inode);
279 if (inode->i_sb->s_root &&
280 inode != inode->i_sb->s_root->d_inode &&
281 (bits & MDS_INODELOCK_LOOKUP))
282 ll_invalidate_aliases(inode);
293 __u32 ll_i2suppgid(struct inode *i)
295 if (cfs_curproc_is_in_groups(i->i_gid))
296 return (__u32)i->i_gid;
301 /* Pack the required supplementary groups into the supplied groups array.
302 * If we don't need to use the groups from the target inode(s) then we
303 * instead pack one or more groups from the user's supplementary group
304 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
305 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
312 LASSERT(suppgids != NULL);
314 suppgids[0] = ll_i2suppgid(i1);
317 suppgids[1] = ll_i2suppgid(i2);
322 for (i = 0; i < current_ngroups; i++) {
323 if (suppgids[0] == -1) {
324 if (current_groups[i] != suppgids[1])
325 suppgids[0] = current_groups[i];
328 if (suppgids[1] == -1) {
329 if (current_groups[i] != suppgids[0])
330 suppgids[1] = current_groups[i];
339 * try to reuse three types of dentry:
340 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
341 * by concurrent .revalidate).
342 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
343 * be cleared by others calling d_lustre_revalidate).
344 * 3. DISCONNECTED alias.
346 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
348 struct dentry *alias, *discon_alias, *invalid_alias;
349 struct ll_d_hlist_node *p;
351 if (ll_d_hlist_empty(&inode->i_dentry))
354 discon_alias = invalid_alias = NULL;
356 ll_lock_dcache(inode);
357 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
358 LASSERT(alias != dentry);
360 spin_lock(&alias->d_lock);
361 if (alias->d_flags & DCACHE_DISCONNECTED)
362 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
363 discon_alias = alias;
364 else if (alias->d_parent == dentry->d_parent &&
365 alias->d_name.hash == dentry->d_name.hash &&
366 alias->d_name.len == dentry->d_name.len &&
367 memcmp(alias->d_name.name, dentry->d_name.name,
368 dentry->d_name.len) == 0)
369 invalid_alias = alias;
370 spin_unlock(&alias->d_lock);
375 alias = invalid_alias ?: discon_alias ?: NULL;
377 spin_lock(&alias->d_lock);
379 spin_unlock(&alias->d_lock);
381 ll_unlock_dcache(inode);
387 * Similar to d_splice_alias(), but lustre treats invalid alias
388 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
390 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
395 new = ll_find_alias(inode, de);
397 ll_dops_init(new, 1, 1);
401 "Reuse dentry %p inode %p refc %d flags %#x\n",
402 new, new->d_inode, d_refcount(new), new->d_flags);
406 ll_dops_init(de, 1, 1);
407 __d_lustre_invalidate(de);
409 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
410 de, de->d_inode, d_refcount(de), de->d_flags);
414 int ll_lookup_it_finish(struct ptlrpc_request *request,
415 struct lookup_intent *it, void *data)
417 struct it_cb_data *icbd = data;
418 struct dentry **de = icbd->icbd_childp;
419 struct inode *parent = icbd->icbd_parent;
420 struct inode *inode = NULL;
425 /* NB 1 request reference will be taken away by ll_intent_lock()
427 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
428 it->d.lustre.it_disposition);
429 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
430 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
434 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
436 /* We used to query real size from OSTs here, but actually
437 this is not needed. For stat() calls size would be updated
438 from subsequent do_revalidate()->ll_inode_revalidate_it() in
440 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
441 Everybody else who needs correct file size would call
442 ll_glimpse_size or some equivalent themselves anyway.
443 Also see bug 7198. */
446 /* Only hash *de if it is unhashed (new dentry).
447 * Atoimc_open may passin hashed dentries for open.
450 *de = ll_splice_alias(inode, *de);
452 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
453 /* we have lookup look - unhide dentry */
454 if (bits & MDS_INODELOCK_LOOKUP)
455 d_lustre_revalidate(*de);
456 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
457 /* If file created on server, don't depend on parent UPDATE
458 * lock to unhide it. It is left hidden and next lookup can
459 * find it in ll_splice_alias.
461 /* Check that parent has UPDATE lock. */
462 struct lookup_intent parent_it = {
464 .d.lustre.it_lock_handle = 0 };
466 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
467 &ll_i2info(parent)->lli_fid, NULL)) {
468 d_lustre_revalidate(*de);
469 ll_intent_release(&parent_it);
476 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
477 struct lookup_intent *it, int lookup_flags)
479 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
480 struct dentry *save = dentry, *retval;
481 struct ptlrpc_request *req = NULL;
482 struct md_op_data *op_data;
483 struct it_cb_data icbd;
488 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
489 RETURN(ERR_PTR(-ENAMETOOLONG));
491 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
492 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
493 parent->i_generation, parent, LL_IT2STR(it));
495 if (d_mountpoint(dentry))
496 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
498 ll_frob_intent(&it, &lookup_it);
500 /* As do_lookup is called before follow_mount, root dentry may be left
501 * not valid, revalidate it here. */
502 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
503 (it->it_op & (IT_OPEN | IT_CREAT))) {
504 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
505 MDS_INODELOCK_LOOKUP);
510 if (it->it_op == IT_GETATTR) {
511 rc = ll_statahead_enter(parent, &dentry, 0);
514 GOTO(out, retval = NULL);
515 GOTO(out, retval = dentry);
519 icbd.icbd_childp = &dentry;
520 icbd.icbd_parent = parent;
522 if (it->it_op & IT_CREAT ||
523 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
524 opc = LUSTRE_OPC_CREATE;
526 opc = LUSTRE_OPC_ANY;
528 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
529 dentry->d_name.len, lookup_flags, opc,
532 RETURN((void *)op_data);
534 /* enforce umask if acl disabled or MDS doesn't support umask */
535 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
536 it->it_create_mode &= ~cfs_curproc_umask();
538 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
539 lookup_flags, &req, ll_md_blocking_ast, 0);
540 ll_finish_md_op_data(op_data);
542 GOTO(out, retval = ERR_PTR(rc));
544 rc = ll_lookup_it_finish(req, it, &icbd);
546 ll_intent_release(it);
547 GOTO(out, retval = ERR_PTR(rc));
550 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
551 !S_ISREG(dentry->d_inode->i_mode) &&
552 !S_ISDIR(dentry->d_inode->i_mode)) {
553 ll_release_openhandle(dentry, it);
555 ll_lookup_finish_locks(it, dentry);
558 GOTO(out, retval = NULL);
560 GOTO(out, retval = dentry);
563 ptlrpc_req_finished(req);
564 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
565 ll_statahead_mark(parent, dentry);
569 #ifdef HAVE_IOP_ATOMIC_OPEN
570 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
573 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
576 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),flags=%u\n",
577 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
578 parent->i_generation, parent, flags);
580 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
581 if ((flags & LOOKUP_CREATE ) && !(flags & LOOKUP_OPEN)) {
582 ll_dops_init(dentry, 1, 1);
583 __d_lustre_invalidate(dentry);
588 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
592 de = ll_lookup_it(parent, dentry, itp, 0);
595 ll_intent_release(itp);
601 * For cached negative dentry and new dentry, handle lookup/create/open
604 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
605 struct file *file, unsigned open_flags,
606 umode_t mode, int *opened)
608 struct lookup_intent *it;
610 long long lookup_flags = LOOKUP_OPEN;
614 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),file %p,"
615 "open_flags %x,mode %x opened %d\n",
616 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
617 dir->i_generation, dir, file, open_flags, mode, *opened);
619 OBD_ALLOC(it, sizeof(*it));
625 it->it_op |= IT_CREAT;
626 lookup_flags |= LOOKUP_CREATE;
628 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
629 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
631 /* Dentry added to dcache tree in ll_lookup_it */
632 de = ll_lookup_it(dir, dentry, it, lookup_flags);
639 if (it_disposition(it, DISP_OPEN_CREATE)) {
640 /* Dentry instantiated in ll_create_it. */
641 rc = ll_create_it(dir, dentry, mode, it);
643 /* We dget in ll_splice_alias. */
649 *opened |= FILE_CREATED;
651 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
653 if (S_ISFIFO(dentry->d_inode->i_mode)) {
654 /* We cannot call open here as it would
657 if (it_disposition(it, DISP_ENQ_OPEN_REF))
659 (struct ptlrpc_request *)
660 it->d.lustre.it_data);
661 rc = finish_no_open(file, de);
663 file->private_data = it;
664 rc = finish_open(file, dentry, NULL, opened);
665 /* We dget in ll_splice_alias. finish_open takes
666 * care of dget for fd open.
672 rc = finish_no_open(file, de);
677 ll_intent_release(it);
678 OBD_FREE(it, sizeof(*it));
683 #else /* !HAVE_IOP_ATOMIC_OPEN */
684 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
687 struct lookup_intent *it;
689 OBD_ALLOC(it, sizeof(*it));
691 return ERR_PTR(-ENOMEM);
693 if (lookup_flags & LOOKUP_OPEN) {
695 if (lookup_flags & LOOKUP_CREATE)
696 it->it_op |= IT_CREAT;
697 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
698 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
700 it->it_op = IT_GETATTR;
706 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
707 struct nameidata *nd)
712 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
713 struct lookup_intent *it;
715 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
716 it = ll_d2d(dentry)->lld_it;
717 ll_d2d(dentry)->lld_it = NULL;
719 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
720 /* We are sure this is new dentry, so we need to create
721 our private data and set the dentry ops */
722 ll_dops_init(dentry, 1, 1);
723 __d_lustre_invalidate(dentry);
727 it = ll_convert_intent(&nd->intent.open, nd->flags);
729 RETURN((struct dentry *)it);
732 de = ll_lookup_it(parent, dentry, it, nd->flags);
735 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
736 if (dentry->d_inode &&
737 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
738 if (S_ISFIFO(dentry->d_inode->i_mode)) {
739 // We cannot call open here as it would
742 (struct ptlrpc_request *)
743 it->d.lustre.it_data);
747 nd->intent.open.file->private_data = it;
748 filp = lookup_instantiate_filp(nd,
754 de = (struct dentry *)filp;
757 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
758 // XXX This can only reliably work on assumption
759 // that there are NO hashed negative dentries.
760 ll_d2d(dentry)->lld_it = it;
761 it = NULL; /* Will be freed in ll_create_nd */
762 /* We absolutely depend on ll_create_nd to be
763 * called to not leak this intent and possible
764 * data attached to it */
769 ll_intent_release(it);
770 OBD_FREE(it, sizeof(*it));
773 de = ll_lookup_it(parent, dentry, NULL, 0);
778 #endif /* HAVE_IOP_ATOMIC_OPEN */
780 /* We depend on "mode" being set with the proper file type/umask by now */
781 static struct inode *ll_create_node(struct inode *dir, const char *name,
782 int namelen, const void *data, int datalen,
783 int mode, __u64 extra,
784 struct lookup_intent *it)
786 struct inode *inode = NULL;
787 struct ptlrpc_request *request = NULL;
788 struct ll_sb_info *sbi = ll_i2sbi(dir);
792 LASSERT(it && it->d.lustre.it_disposition);
794 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
795 request = it->d.lustre.it_data;
796 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
797 rc = ll_prep_inode(&inode, request, dir->i_sb);
799 GOTO(out, inode = ERR_PTR(rc));
801 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
803 /* We asked for a lock on the directory, but were granted a
804 * lock on the inode. Since we finally have an inode pointer,
805 * stuff it in the lock. */
806 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
807 inode, inode->i_ino, inode->i_generation);
808 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
811 ptlrpc_req_finished(request);
816 * By the time this is called, we already have created the directory cache
817 * entry for the new file, but it is so far negative - it has no inode.
819 * We defer creating the OBD object(s) until open, to keep the intent and
820 * non-intent code paths similar, and also because we do not have the MDS
821 * inode number before calling ll_create_node() (which is needed for LOV),
822 * so we would need to do yet another RPC to the MDS to store the LOV EA
823 * data on the MDS. If needed, we would pass the PACKED lmm as data and
824 * lmm_size in datalen (the MDS still has code which will handle that).
826 * If the create succeeds, we fill in the inode information
827 * with d_instantiate().
829 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
830 struct lookup_intent *it)
836 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
837 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
838 dir->i_generation, dir, LL_IT2STR(it));
840 rc = it_open_error(DISP_OPEN_CREATE, it);
844 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
845 NULL, 0, mode, 0, it);
847 RETURN(PTR_ERR(inode));
849 d_instantiate(dentry, inode);
853 static void ll_update_times(struct ptlrpc_request *request,
856 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
860 if (body->valid & OBD_MD_FLMTIME &&
861 body->mtime > LTIME_S(inode->i_mtime)) {
862 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
863 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
864 LTIME_S(inode->i_mtime) = body->mtime;
866 if (body->valid & OBD_MD_FLCTIME &&
867 body->ctime > LTIME_S(inode->i_ctime))
868 LTIME_S(inode->i_ctime) = body->ctime;
871 static int ll_new_node(struct inode *dir, struct qstr *name,
872 const char *tgt, int mode, int rdev,
873 struct dentry *dchild, __u32 opc)
875 struct ptlrpc_request *request = NULL;
876 struct md_op_data *op_data;
877 struct inode *inode = NULL;
878 struct ll_sb_info *sbi = ll_i2sbi(dir);
883 if (unlikely(tgt != NULL))
884 tgt_len = strlen(tgt) + 1;
886 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
887 name->len, 0, opc, NULL);
889 GOTO(err_exit, err = PTR_ERR(op_data));
891 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
892 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
893 cfs_curproc_cap_pack(), rdev, &request);
894 ll_finish_md_op_data(op_data);
898 ll_update_times(request, dir);
901 err = ll_prep_inode(&inode, request, dchild->d_sb);
905 d_instantiate(dchild, inode);
909 ptlrpc_req_finished(request);
914 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
915 unsigned rdev, struct dentry *dchild)
920 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
921 name->len, name->name, dir->i_ino, dir->i_generation, dir,
924 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
925 mode &= ~cfs_curproc_umask();
927 switch (mode & S_IFMT) {
929 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
935 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
946 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
951 #ifdef HAVE_IOP_ATOMIC_OPEN
953 * Plain create. Intent create is handled in atomic_open.
955 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
956 umode_t mode, bool want_excl)
960 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),"
961 "flags=%u, excl=%d\n",
962 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
963 dir->i_generation, dir, mode, want_excl);
965 rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
967 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
969 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
970 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
974 #else /* !HAVE_IOP_ATOMIC_OPEN */
975 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
976 ll_umode_t mode, struct nameidata *nd)
978 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
982 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
984 ll_d2d(dentry)->lld_it = NULL;
986 /* Was there an error? Propagate it! */
987 if (it->d.lustre.it_status) {
988 rc = it->d.lustre.it_status;
992 rc = ll_create_it(dir, dentry, mode, it);
993 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
996 nd->intent.open.file->private_data = it;
997 filp = lookup_instantiate_filp(nd, dentry, NULL);
1003 ll_intent_release(it);
1004 OBD_FREE(it, sizeof(*it));
1007 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1011 #endif /* HAVE_IOP_ATOMIC_OPEN */
1013 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1014 const char *tgt, struct dentry *dchild)
1019 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
1020 name->len, name->name, dir->i_ino, dir->i_generation,
1023 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1024 0, dchild, LUSTRE_OPC_SYMLINK);
1027 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1032 static int ll_link_generic(struct inode *src, struct inode *dir,
1033 struct qstr *name, struct dentry *dchild)
1035 struct ll_sb_info *sbi = ll_i2sbi(dir);
1036 struct ptlrpc_request *request = NULL;
1037 struct md_op_data *op_data;
1042 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1043 src->i_ino, src->i_generation, src, dir->i_ino,
1044 dir->i_generation, dir, name->len, name->name);
1046 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1047 0, LUSTRE_OPC_ANY, NULL);
1048 if (IS_ERR(op_data))
1049 RETURN(PTR_ERR(op_data));
1051 err = md_link(sbi->ll_md_exp, op_data, &request);
1052 ll_finish_md_op_data(op_data);
1056 ll_update_times(request, dir);
1057 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1060 ptlrpc_req_finished(request);
1064 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1065 int mode, struct dentry *dchild)
1071 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1072 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1074 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1075 mode &= ~cfs_curproc_umask();
1076 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1077 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1080 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1085 /* Try to find the child dentry by its name.
1086 If found, put the result fid into @fid. */
1087 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1090 struct dentry *parent, *child;
1092 parent = ll_d_hlist_entry(dir->i_dentry, struct dentry, d_alias);
1093 child = d_lookup(parent, name);
1096 *fid = *ll_inode2fid(child->d_inode);
1101 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1102 struct dentry *dchild, struct qstr *name)
1104 struct ptlrpc_request *request = NULL;
1105 struct md_op_data *op_data;
1109 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1110 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1112 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1115 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1116 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1117 if (IS_ERR(op_data))
1118 RETURN(PTR_ERR(op_data));
1120 ll_get_child_fid(dir, name, &op_data->op_fid3);
1121 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1122 ll_finish_md_op_data(op_data);
1124 ll_update_times(request, dir);
1125 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1128 ptlrpc_req_finished(request);
1132 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1134 struct mdt_body *body;
1135 struct lov_mds_md *eadata;
1136 struct lov_stripe_md *lsm = NULL;
1137 struct obd_trans_info oti = { 0 };
1139 struct obd_capa *oc = NULL;
1143 /* req is swabbed so this is safe */
1144 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1145 if (!(body->valid & OBD_MD_FLEASIZE))
1148 if (body->eadatasize == 0) {
1149 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1150 GOTO(out, rc = -EPROTO);
1153 /* The MDS sent back the EA because we unlinked the last reference
1154 * to this file. Use this EA to unlink the objects on the OST.
1155 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1156 * check it is complete and sensible. */
1157 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1159 LASSERT(eadata != NULL);
1161 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1163 CERROR("obd_unpackmd: %d\n", rc);
1166 LASSERT(rc >= sizeof(*lsm));
1170 GOTO(out_free_memmd, rc = -ENOMEM);
1172 oa->o_id = lsm->lsm_object_id;
1173 oa->o_seq = lsm->lsm_object_seq;
1174 oa->o_mode = body->mode & S_IFMT;
1175 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1177 if (body->valid & OBD_MD_FLCOOKIE) {
1178 oa->o_valid |= OBD_MD_FLCOOKIE;
1179 oti.oti_logcookies =
1180 req_capsule_server_sized_get(&request->rq_pill,
1182 sizeof(struct llog_cookie) *
1183 lsm->lsm_stripe_count);
1184 if (oti.oti_logcookies == NULL) {
1185 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1186 body->valid &= ~OBD_MD_FLCOOKIE;
1190 if (body->valid & OBD_MD_FLOSSCAPA) {
1191 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1193 GOTO(out_free_memmd, rc);
1196 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1197 ll_i2mdexp(dir), oc);
1201 CERROR("obd destroy objid "LPX64" error %d\n",
1202 lsm->lsm_object_id, rc);
1204 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1209 /* ll_unlink_generic() doesn't update the inode with the new link count.
1210 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1211 * is any lock existing. They will recycle dentries and inodes based upon locks
1213 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1214 struct dentry *dchild, struct qstr *name)
1216 struct ptlrpc_request *request = NULL;
1217 struct md_op_data *op_data;
1220 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1221 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1224 * XXX: unlink bind mountpoint maybe call to here,
1225 * just check it as vfs_unlink does.
1227 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1230 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1231 name->len, 0, LUSTRE_OPC_ANY, NULL);
1232 if (IS_ERR(op_data))
1233 RETURN(PTR_ERR(op_data));
1235 ll_get_child_fid(dir, name, &op_data->op_fid3);
1236 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1237 ll_finish_md_op_data(op_data);
1241 ll_update_times(request, dir);
1242 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1244 rc = ll_objects_destroy(request, dir);
1246 ptlrpc_req_finished(request);
1250 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1251 struct dentry *src_dchild, struct qstr *src_name,
1252 struct inode *tgt, struct dentry *tgt_dparent,
1253 struct dentry *tgt_dchild, struct qstr *tgt_name)
1255 struct ptlrpc_request *request = NULL;
1256 struct ll_sb_info *sbi = ll_i2sbi(src);
1257 struct md_op_data *op_data;
1260 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1261 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1262 src->i_ino, src->i_generation, src, tgt_name->len,
1263 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1265 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1266 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1269 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1270 LUSTRE_OPC_ANY, NULL);
1271 if (IS_ERR(op_data))
1272 RETURN(PTR_ERR(op_data));
1274 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1275 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1276 err = md_rename(sbi->ll_md_exp, op_data,
1277 src_name->name, src_name->len,
1278 tgt_name->name, tgt_name->len, &request);
1279 ll_finish_md_op_data(op_data);
1281 ll_update_times(request, src);
1282 ll_update_times(request, tgt);
1283 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1284 err = ll_objects_destroy(request, src);
1287 ptlrpc_req_finished(request);
1292 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1295 return ll_mknod_generic(dir, &dchild->d_name, mode,
1296 old_encode_dev(rdev), dchild);
1299 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1301 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1304 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1306 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1309 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1311 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1314 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1315 const char *oldname)
1317 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1320 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1321 struct dentry *new_dentry)
1323 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1327 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1328 struct inode *new_dir, struct dentry *new_dentry)
1331 err = ll_rename_generic(old_dir, NULL,
1332 old_dentry, &old_dentry->d_name,
1333 new_dir, NULL, new_dentry,
1334 &new_dentry->d_name);
1336 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1337 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1339 d_move(old_dentry, new_dentry);
1344 struct inode_operations ll_dir_inode_operations = {
1346 #ifdef HAVE_IOP_ATOMIC_OPEN
1347 .atomic_open = ll_atomic_open,
1349 .lookup = ll_lookup_nd,
1350 .create = ll_create_nd,
1351 /* We need all these non-raw things for NFSD, to not patch it. */
1352 .unlink = ll_unlink,
1355 .symlink = ll_symlink,
1357 .rename = ll_rename,
1358 .setattr = ll_setattr,
1359 .getattr = ll_getattr,
1360 .permission = ll_inode_permission,
1361 .setxattr = ll_setxattr,
1362 .getxattr = ll_getxattr,
1363 .listxattr = ll_listxattr,
1364 .removexattr = ll_removexattr,
1365 #ifdef HAVE_IOP_GET_ACL
1366 .get_acl = ll_get_acl,
1370 struct inode_operations ll_special_inode_operations = {
1371 .setattr = ll_setattr,
1372 .getattr = ll_getattr,
1373 .permission = ll_inode_permission,
1374 .setxattr = ll_setxattr,
1375 .getxattr = ll_getxattr,
1376 .listxattr = ll_listxattr,
1377 .removexattr = ll_removexattr,
1378 #ifdef HAVE_IOP_GET_ACL
1379 .get_acl = ll_get_acl,