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, 2013, 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) {
155 "%s: apply lsm %p to inode "DFID".\n",
156 ll_get_fsname(sb, NULL, 0), md->lsm,
157 PFID(ll_inode2fid(inode)));
158 rc = cl_file_inode_init(inode, md);
161 make_bad_inode(inode);
162 unlock_new_inode(inode);
166 unlock_new_inode(inode);
167 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
168 ll_update_inode(inode, md);
169 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p)\n",
170 PFID(&md->body->fid1), inode);
175 static void ll_invalidate_negative_children(struct inode *dir)
177 struct dentry *dentry, *tmp_subdir;
178 DECLARE_LL_D_HLIST_NODE_PTR(p);
181 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
182 spin_lock(&dentry->d_lock);
183 if (!list_empty(&dentry->d_subdirs)) {
184 struct dentry *child;
186 list_for_each_entry_safe(child, tmp_subdir,
189 if (child->d_inode == NULL)
190 d_lustre_invalidate(child, 1);
193 spin_unlock(&dentry->d_lock);
195 ll_unlock_dcache(dir);
198 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
199 void *data, int flag)
202 struct lustre_handle lockh;
206 case LDLM_CB_BLOCKING:
207 ldlm_lock2handle(lock, &lockh);
208 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
210 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
214 case LDLM_CB_CANCELING: {
215 struct inode *inode = ll_inode_from_resource_lock(lock);
216 struct ll_inode_info *lli;
217 __u64 bits = lock->l_policy_data.l_inodebits.bits;
219 ldlm_mode_t mode = lock->l_req_mode;
221 /* Inode is set to lock->l_resource->lr_lvb_inode
222 * for mdc - bug 24555 */
223 LASSERT(lock->l_ast_data == NULL);
225 /* Invalidate all dentries associated with this inode */
229 LASSERT(ldlm_is_canceling(lock));
231 if (bits & MDS_INODELOCK_XATTR)
232 ll_xattr_cache_destroy(inode);
234 /* For OPEN locks we differentiate between lock modes
235 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
236 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
237 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
238 ll_have_md_lock(inode, &bits, LCK_MINMODE);
240 if (bits & MDS_INODELOCK_OPEN)
241 ll_have_md_lock(inode, &bits, mode);
243 fid = ll_inode2fid(inode);
244 if (!fid_res_name_eq(fid, &lock->l_resource->lr_name))
245 LDLM_ERROR(lock, "data mismatch with object "
246 DFID" (%p)", PFID(fid), inode);
248 if (bits & MDS_INODELOCK_OPEN) {
250 switch (lock->l_req_mode) {
261 CERROR("%s: unexpected lock mode for OPEN lock"
262 " %d, inode "DFID"\n",
263 ll_get_fsname(inode->i_sb, NULL, 0),
265 PFID(ll_inode2fid(inode)));
267 ll_md_real_close(inode, flags);
270 lli = ll_i2info(inode);
271 if (bits & MDS_INODELOCK_LAYOUT) {
272 struct cl_object_conf conf = { { 0 } };
274 conf.coc_opc = OBJECT_CONF_INVALIDATE;
275 conf.coc_inode = inode;
276 rc = ll_layout_conf(inode, &conf);
278 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
281 if (bits & MDS_INODELOCK_UPDATE) {
282 spin_lock(&lli->lli_lock);
283 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
284 spin_unlock(&lli->lli_lock);
287 if (S_ISDIR(inode->i_mode) &&
288 (bits & MDS_INODELOCK_UPDATE)) {
289 CDEBUG(D_INODE, "invalidating inode "DFID"\n",
290 PFID(ll_inode2fid(inode)));
291 truncate_inode_pages(inode->i_mapping, 0);
292 ll_invalidate_negative_children(inode);
295 if (inode->i_sb->s_root &&
296 inode != inode->i_sb->s_root->d_inode &&
297 (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)))
298 ll_invalidate_aliases(inode);
309 __u32 ll_i2suppgid(struct inode *i)
311 if (in_group_p(i->i_gid))
312 return (__u32)i->i_gid;
317 /* Pack the required supplementary groups into the supplied groups array.
318 * If we don't need to use the groups from the target inode(s) then we
319 * instead pack one or more groups from the user's supplementary group
320 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
321 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
328 LASSERT(suppgids != NULL);
330 suppgids[0] = ll_i2suppgid(i1);
333 suppgids[1] = ll_i2suppgid(i2);
338 for (i = 0; i < current_ngroups; i++) {
339 if (suppgids[0] == -1) {
340 if (current_groups[i] != suppgids[1])
341 suppgids[0] = current_groups[i];
344 if (suppgids[1] == -1) {
345 if (current_groups[i] != suppgids[0])
346 suppgids[1] = current_groups[i];
355 * try to reuse three types of dentry:
356 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
357 * by concurrent .revalidate).
358 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
359 * be cleared by others calling d_lustre_revalidate).
360 * 3. DISCONNECTED alias.
362 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
364 struct dentry *alias, *discon_alias, *invalid_alias;
365 DECLARE_LL_D_HLIST_NODE_PTR(p);
367 if (ll_d_hlist_empty(&inode->i_dentry))
370 discon_alias = invalid_alias = NULL;
372 ll_lock_dcache(inode);
373 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
374 LASSERT(alias != dentry);
376 spin_lock(&alias->d_lock);
377 if (alias->d_flags & DCACHE_DISCONNECTED)
378 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
379 discon_alias = alias;
380 else if (alias->d_parent == dentry->d_parent &&
381 alias->d_name.hash == dentry->d_name.hash &&
382 alias->d_name.len == dentry->d_name.len &&
383 memcmp(alias->d_name.name, dentry->d_name.name,
384 dentry->d_name.len) == 0)
385 invalid_alias = alias;
386 spin_unlock(&alias->d_lock);
391 alias = invalid_alias ?: discon_alias ?: NULL;
393 spin_lock(&alias->d_lock);
395 spin_unlock(&alias->d_lock);
397 ll_unlock_dcache(inode);
403 * Similar to d_splice_alias(), but lustre treats invalid alias
404 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
406 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
412 new = ll_find_alias(inode, de);
422 "Reuse dentry %p inode %p refc %d flags %#x\n",
423 new, new->d_inode, d_refcount(new), new->d_flags);
431 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
432 de, de->d_inode, d_refcount(de), de->d_flags);
436 int ll_lookup_it_finish(struct ptlrpc_request *request,
437 struct lookup_intent *it, void *data)
439 struct it_cb_data *icbd = data;
440 struct dentry **de = icbd->icbd_childp;
441 struct inode *parent = icbd->icbd_parent;
442 struct inode *inode = NULL;
447 /* NB 1 request reference will be taken away by ll_intent_lock()
449 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
450 it->d.lustre.it_disposition);
451 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
452 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
456 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
458 /* We used to query real size from OSTs here, but actually
459 this is not needed. For stat() calls size would be updated
460 from subsequent do_revalidate()->ll_inode_revalidate_it() in
462 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
463 Everybody else who needs correct file size would call
464 ll_glimpse_size or some equivalent themselves anyway.
465 Also see bug 7198. */
468 /* Only hash *de if it is unhashed (new dentry).
469 * Atoimc_open may passin hashed dentries for open.
471 if (d_unhashed(*de)) {
472 struct dentry *alias;
474 alias = ll_splice_alias(inode, *de);
476 RETURN(PTR_ERR(alias));
480 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
481 /* we have lookup look - unhide dentry */
482 if (bits & MDS_INODELOCK_LOOKUP)
483 d_lustre_revalidate(*de);
484 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
485 /* If file created on server, don't depend on parent UPDATE
486 * lock to unhide it. It is left hidden and next lookup can
487 * find it in ll_splice_alias.
489 /* Check that parent has UPDATE lock. */
490 struct lookup_intent parent_it = {
492 .d.lustre.it_lock_handle = 0 };
494 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
495 &ll_i2info(parent)->lli_fid, NULL)) {
496 d_lustre_revalidate(*de);
497 ll_intent_release(&parent_it);
504 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
505 struct lookup_intent *it, int lookup_flags)
507 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
508 struct dentry *save = dentry, *retval;
509 struct ptlrpc_request *req = NULL;
510 struct md_op_data *op_data;
511 struct it_cb_data icbd;
516 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
517 RETURN(ERR_PTR(-ENAMETOOLONG));
519 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
520 dentry->d_name.len, dentry->d_name.name,
521 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
523 if (d_mountpoint(dentry))
524 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
526 ll_frob_intent(&it, &lookup_it);
528 if (it->it_op == IT_GETATTR) {
529 rc = ll_statahead_enter(parent, &dentry, 0);
532 GOTO(out, retval = NULL);
533 GOTO(out, retval = dentry);
537 icbd.icbd_childp = &dentry;
538 icbd.icbd_parent = parent;
540 if (it->it_op & IT_CREAT)
541 opc = LUSTRE_OPC_CREATE;
543 opc = LUSTRE_OPC_ANY;
545 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
546 dentry->d_name.len, lookup_flags, opc,
549 RETURN((void *)op_data);
551 /* enforce umask if acl disabled or MDS doesn't support umask */
552 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
553 it->it_create_mode &= ~current_umask();
555 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
556 lookup_flags, &req, ll_md_blocking_ast, 0);
557 ll_finish_md_op_data(op_data);
559 GOTO(out, retval = ERR_PTR(rc));
561 rc = ll_lookup_it_finish(req, it, &icbd);
563 ll_intent_release(it);
564 GOTO(out, retval = ERR_PTR(rc));
567 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
568 !S_ISREG(dentry->d_inode->i_mode) &&
569 !S_ISDIR(dentry->d_inode->i_mode)) {
570 ll_release_openhandle(dentry, it);
572 ll_lookup_finish_locks(it, dentry);
575 GOTO(out, retval = NULL);
577 GOTO(out, retval = dentry);
580 ptlrpc_req_finished(req);
581 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
582 ll_statahead_mark(parent, dentry);
586 #ifdef HAVE_IOP_ATOMIC_OPEN
587 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
590 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
593 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
594 dentry->d_name.len, dentry->d_name.name,
595 PFID(ll_inode2fid(parent)), parent, flags);
597 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
598 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
601 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
605 de = ll_lookup_it(parent, dentry, itp, 0);
608 ll_intent_release(itp);
614 * For cached negative dentry and new dentry, handle lookup/create/open
617 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
618 struct file *file, unsigned open_flags,
619 umode_t mode, int *opened)
621 struct lookup_intent *it;
623 long long lookup_flags = LOOKUP_OPEN;
627 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
628 "open_flags %x, mode %x opened %d\n",
629 dentry->d_name.len, dentry->d_name.name,
630 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
632 OBD_ALLOC(it, sizeof(*it));
637 if (open_flags & O_CREAT) {
638 it->it_op |= IT_CREAT;
639 lookup_flags |= LOOKUP_CREATE;
641 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
642 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
644 /* Dentry added to dcache tree in ll_lookup_it */
645 de = ll_lookup_it(dir, dentry, it, lookup_flags);
652 if (it_disposition(it, DISP_OPEN_CREATE)) {
653 /* Dentry instantiated in ll_create_it. */
654 rc = ll_create_it(dir, dentry, mode, it);
656 /* We dget in ll_splice_alias. */
662 *opened |= FILE_CREATED;
664 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
666 if (S_ISFIFO(dentry->d_inode->i_mode)) {
667 /* We cannot call open here as it would
670 if (it_disposition(it, DISP_ENQ_OPEN_REF))
672 (struct ptlrpc_request *)
673 it->d.lustre.it_data);
674 rc = finish_no_open(file, de);
676 file->private_data = it;
677 rc = finish_open(file, dentry, NULL, opened);
678 /* We dget in ll_splice_alias. finish_open takes
679 * care of dget for fd open.
685 rc = finish_no_open(file, de);
690 ll_intent_release(it);
691 OBD_FREE(it, sizeof(*it));
696 #else /* !HAVE_IOP_ATOMIC_OPEN */
697 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
700 struct lookup_intent *it;
702 OBD_ALLOC(it, sizeof(*it));
704 return ERR_PTR(-ENOMEM);
706 if (lookup_flags & LOOKUP_OPEN) {
708 if (lookup_flags & LOOKUP_CREATE)
709 it->it_op |= IT_CREAT;
710 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
711 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
713 it->it_op = IT_GETATTR;
719 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
720 struct nameidata *nd)
725 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
726 struct lookup_intent *it;
728 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
729 it = ll_d2d(dentry)->lld_it;
730 ll_d2d(dentry)->lld_it = NULL;
732 if ((nd->flags & LOOKUP_CREATE) &&
733 !(nd->flags & LOOKUP_OPEN))
736 it = ll_convert_intent(&nd->intent.open, nd->flags);
738 RETURN((struct dentry *)it);
741 de = ll_lookup_it(parent, dentry, it, nd->flags);
744 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
745 if (dentry->d_inode &&
746 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
747 if (S_ISFIFO(dentry->d_inode->i_mode)) {
748 // We cannot call open here as it would
751 (struct ptlrpc_request *)
752 it->d.lustre.it_data);
756 nd->intent.open.file->private_data = it;
757 filp = lookup_instantiate_filp(nd,
763 de = (struct dentry *)filp;
766 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
767 // XXX This can only reliably work on assumption
768 // that there are NO hashed negative dentries.
769 ll_d2d(dentry)->lld_it = it;
770 it = NULL; /* Will be freed in ll_create_nd */
771 /* We absolutely depend on ll_create_nd to be
772 * called to not leak this intent and possible
773 * data attached to it */
778 ll_intent_release(it);
779 OBD_FREE(it, sizeof(*it));
782 de = ll_lookup_it(parent, dentry, NULL, 0);
787 #endif /* HAVE_IOP_ATOMIC_OPEN */
789 /* We depend on "mode" being set with the proper file type/umask by now */
790 static struct inode *ll_create_node(struct inode *dir, const char *name,
791 int namelen, const void *data, int datalen,
792 int mode, __u64 extra,
793 struct lookup_intent *it)
795 struct inode *inode = NULL;
796 struct ptlrpc_request *request = NULL;
797 struct ll_sb_info *sbi = ll_i2sbi(dir);
801 LASSERT(it && it->d.lustre.it_disposition);
803 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
804 request = it->d.lustre.it_data;
805 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
806 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
808 GOTO(out, inode = ERR_PTR(rc));
810 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
812 /* We asked for a lock on the directory, but were granted a
813 * lock on the inode. Since we finally have an inode pointer,
814 * stuff it in the lock. */
815 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
816 PFID(ll_inode2fid(inode)), inode);
817 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
820 ptlrpc_req_finished(request);
825 * By the time this is called, we already have created the directory cache
826 * entry for the new file, but it is so far negative - it has no inode.
828 * We defer creating the OBD object(s) until open, to keep the intent and
829 * non-intent code paths similar, and also because we do not have the MDS
830 * inode number before calling ll_create_node() (which is needed for LOV),
831 * so we would need to do yet another RPC to the MDS to store the LOV EA
832 * data on the MDS. If needed, we would pass the PACKED lmm as data and
833 * lmm_size in datalen (the MDS still has code which will handle that).
835 * If the create succeeds, we fill in the inode information
836 * with d_instantiate().
838 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
839 struct lookup_intent *it)
845 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
846 dentry->d_name.len, dentry->d_name.name,
847 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
849 rc = it_open_error(DISP_OPEN_CREATE, it);
853 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
854 NULL, 0, mode, 0, it);
856 RETURN(PTR_ERR(inode));
858 if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
859 ll_i2info(inode)->lli_volatile = true;
861 d_instantiate(dentry, inode);
865 static void ll_update_times(struct ptlrpc_request *request,
868 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
872 if (body->valid & OBD_MD_FLMTIME &&
873 body->mtime > LTIME_S(inode->i_mtime)) {
874 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
875 "\n", PFID(ll_inode2fid(inode)),
876 LTIME_S(inode->i_mtime), body->mtime);
877 LTIME_S(inode->i_mtime) = body->mtime;
879 if (body->valid & OBD_MD_FLCTIME &&
880 body->ctime > LTIME_S(inode->i_ctime))
881 LTIME_S(inode->i_ctime) = body->ctime;
884 static int ll_new_node(struct inode *dir, struct qstr *name,
885 const char *tgt, int mode, int rdev,
886 struct dentry *dchild, __u32 opc)
888 struct ptlrpc_request *request = NULL;
889 struct md_op_data *op_data;
890 struct inode *inode = NULL;
891 struct ll_sb_info *sbi = ll_i2sbi(dir);
896 if (unlikely(tgt != NULL))
897 tgt_len = strlen(tgt) + 1;
899 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
900 name->len, 0, opc, NULL);
902 GOTO(err_exit, err = PTR_ERR(op_data));
904 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
905 current_fsuid(), current_fsgid(),
906 cfs_curproc_cap_pack(), rdev, &request);
907 ll_finish_md_op_data(op_data);
911 ll_update_times(request, dir);
914 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
918 d_instantiate(dchild, inode);
922 ptlrpc_req_finished(request);
927 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
928 unsigned rdev, struct dentry *dchild)
933 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
934 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
937 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
938 mode &= ~current_umask();
940 switch (mode & S_IFMT) {
942 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
948 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
959 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
964 #ifdef HAVE_IOP_ATOMIC_OPEN
966 * Plain create. Intent create is handled in atomic_open.
968 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
969 umode_t mode, bool want_excl)
973 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
974 "flags=%u, excl=%d\n", dentry->d_name.len,
975 dentry->d_name.name, PFID(ll_inode2fid(dir)),
976 dir, mode, want_excl);
978 rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
980 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
982 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
983 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
987 #else /* !HAVE_IOP_ATOMIC_OPEN */
988 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
989 ll_umode_t mode, struct nameidata *nd)
991 struct ll_dentry_data *lld = ll_d2d(dentry);
992 struct lookup_intent *it = NULL;
999 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1003 /* Was there an error? Propagate it! */
1004 if (it->d.lustre.it_status) {
1005 rc = it->d.lustre.it_status;
1009 rc = ll_create_it(dir, dentry, mode, it);
1010 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1013 nd->intent.open.file->private_data = it;
1014 filp = lookup_instantiate_filp(nd, dentry, NULL);
1020 ll_intent_release(it);
1021 OBD_FREE(it, sizeof(*it));
1024 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1028 #endif /* HAVE_IOP_ATOMIC_OPEN */
1030 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1031 const char *tgt, struct dentry *dchild)
1036 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1037 name->len, name->name, PFID(ll_inode2fid(dir)),
1040 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1041 0, dchild, LUSTRE_OPC_SYMLINK);
1044 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1049 static int ll_link_generic(struct inode *src, struct inode *dir,
1050 struct qstr *name, struct dentry *dchild)
1052 struct ll_sb_info *sbi = ll_i2sbi(dir);
1053 struct ptlrpc_request *request = NULL;
1054 struct md_op_data *op_data;
1058 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID
1059 "(%p), target=%.*s\n", PFID(ll_inode2fid(src)), src,
1060 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1062 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1063 0, LUSTRE_OPC_ANY, NULL);
1064 if (IS_ERR(op_data))
1065 RETURN(PTR_ERR(op_data));
1067 err = md_link(sbi->ll_md_exp, op_data, &request);
1068 ll_finish_md_op_data(op_data);
1072 ll_update_times(request, dir);
1073 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1076 ptlrpc_req_finished(request);
1080 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1081 int mode, struct dentry *dchild)
1087 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1088 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1090 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1091 mode &= ~current_umask();
1092 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1093 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1096 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
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="DFID"(%p)\n",
1110 name->len, name->name, PFID(ll_inode2fid(dir)), 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 if (dchild != NULL && dchild->d_inode != NULL)
1121 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1122 op_data->op_fid2 = op_data->op_fid3;
1123 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1124 ll_finish_md_op_data(op_data);
1126 ll_update_times(request, dir);
1127 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1130 ptlrpc_req_finished(request);
1137 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1139 struct ptlrpc_request *request = NULL;
1140 struct md_op_data *op_data;
1144 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1145 namelen, name, PFID(ll_inode2fid(dir)), dir);
1147 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1148 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1149 if (IS_ERR(op_data))
1150 RETURN(PTR_ERR(op_data));
1151 op_data->op_cli_flags |= CLI_RM_ENTRY;
1152 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1153 ll_finish_md_op_data(op_data);
1155 ll_update_times(request, dir);
1156 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1159 ptlrpc_req_finished(request);
1163 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1165 struct mdt_body *body;
1166 struct lov_mds_md *eadata;
1167 struct lov_stripe_md *lsm = NULL;
1168 struct obd_trans_info oti = { 0 };
1170 struct obd_capa *oc = NULL;
1174 /* req is swabbed so this is safe */
1175 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1176 if (!(body->valid & OBD_MD_FLEASIZE))
1179 if (body->eadatasize == 0) {
1180 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1181 GOTO(out, rc = -EPROTO);
1184 /* The MDS sent back the EA because we unlinked the last reference
1185 * to this file. Use this EA to unlink the objects on the OST.
1186 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1187 * check it is complete and sensible. */
1188 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1190 LASSERT(eadata != NULL);
1192 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1194 CERROR("obd_unpackmd: %d\n", rc);
1197 LASSERT(rc >= sizeof(*lsm));
1201 GOTO(out_free_memmd, rc = -ENOMEM);
1203 oa->o_oi = lsm->lsm_oi;
1204 oa->o_mode = body->mode & S_IFMT;
1205 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1207 if (body->valid & OBD_MD_FLCOOKIE) {
1208 oa->o_valid |= OBD_MD_FLCOOKIE;
1209 oti.oti_logcookies =
1210 req_capsule_server_sized_get(&request->rq_pill,
1212 sizeof(struct llog_cookie) *
1213 lsm->lsm_stripe_count);
1214 if (oti.oti_logcookies == NULL) {
1215 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1216 body->valid &= ~OBD_MD_FLCOOKIE;
1220 if (body->valid & OBD_MD_FLOSSCAPA) {
1221 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1223 GOTO(out_free_memmd, rc);
1226 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1227 ll_i2mdexp(dir), oc);
1230 CERROR("obd destroy objid "DOSTID" error %d\n",
1231 POSTID(&lsm->lsm_oi), rc);
1233 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1239 /* ll_unlink_generic() doesn't update the inode with the new link count.
1240 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1241 * is any lock existing. They will recycle dentries and inodes based upon locks
1243 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1244 struct dentry *dchild, struct qstr *name)
1246 struct ptlrpc_request *request = NULL;
1247 struct md_op_data *op_data;
1250 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1251 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1254 * XXX: unlink bind mountpoint maybe call to here,
1255 * just check it as vfs_unlink does.
1257 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1260 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1261 name->len, 0, LUSTRE_OPC_ANY, NULL);
1262 if (IS_ERR(op_data))
1263 RETURN(PTR_ERR(op_data));
1265 if (dchild != NULL && dchild->d_inode != NULL)
1266 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1268 op_data->op_fid2 = op_data->op_fid3;
1269 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1270 ll_finish_md_op_data(op_data);
1274 ll_update_times(request, dir);
1275 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1277 rc = ll_objects_destroy(request, dir);
1279 ptlrpc_req_finished(request);
1283 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1284 struct dentry *src_dchild, struct qstr *src_name,
1285 struct inode *tgt, struct dentry *tgt_dparent,
1286 struct dentry *tgt_dchild, struct qstr *tgt_name)
1288 struct ptlrpc_request *request = NULL;
1289 struct ll_sb_info *sbi = ll_i2sbi(src);
1290 struct md_op_data *op_data;
1293 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1294 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1295 src_name->len, src_name->name,
1296 PFID(ll_inode2fid(src)), src, tgt_name->len,
1297 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1299 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1300 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1303 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1304 LUSTRE_OPC_ANY, NULL);
1305 if (IS_ERR(op_data))
1306 RETURN(PTR_ERR(op_data));
1308 if (src_dchild != NULL && src_dchild->d_inode != NULL)
1309 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1310 if (tgt_dchild != NULL && tgt_dchild->d_inode != NULL)
1311 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1313 err = md_rename(sbi->ll_md_exp, op_data,
1314 src_name->name, src_name->len,
1315 tgt_name->name, tgt_name->len, &request);
1316 ll_finish_md_op_data(op_data);
1318 ll_update_times(request, src);
1319 ll_update_times(request, tgt);
1320 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1321 err = ll_objects_destroy(request, src);
1324 ptlrpc_req_finished(request);
1329 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1332 return ll_mknod_generic(dir, &dchild->d_name, mode,
1333 old_encode_dev(rdev), dchild);
1336 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1338 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1341 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1343 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1346 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1348 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1351 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1352 const char *oldname)
1354 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1357 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1358 struct dentry *new_dentry)
1360 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1364 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1365 struct inode *new_dir, struct dentry *new_dentry)
1368 err = ll_rename_generic(old_dir, NULL,
1369 old_dentry, &old_dentry->d_name,
1370 new_dir, NULL, new_dentry,
1371 &new_dentry->d_name);
1373 d_move(old_dentry, new_dentry);
1377 struct inode_operations ll_dir_inode_operations = {
1379 #ifdef HAVE_IOP_ATOMIC_OPEN
1380 .atomic_open = ll_atomic_open,
1382 .lookup = ll_lookup_nd,
1383 .create = ll_create_nd,
1384 /* We need all these non-raw things for NFSD, to not patch it. */
1385 .unlink = ll_unlink,
1388 .symlink = ll_symlink,
1390 .rename = ll_rename,
1391 .setattr = ll_setattr,
1392 .getattr = ll_getattr,
1393 .permission = ll_inode_permission,
1394 .setxattr = ll_setxattr,
1395 .getxattr = ll_getxattr,
1396 .listxattr = ll_listxattr,
1397 .removexattr = ll_removexattr,
1398 #ifdef HAVE_IOP_GET_ACL
1399 .get_acl = ll_get_acl,
1403 struct inode_operations ll_special_inode_operations = {
1404 .setattr = ll_setattr,
1405 .getattr = ll_getattr,
1406 .permission = ll_inode_permission,
1407 .setxattr = ll_setxattr,
1408 .getxattr = ll_getxattr,
1409 .listxattr = ll_listxattr,
1410 .removexattr = ll_removexattr,
1411 #ifdef HAVE_IOP_GET_ACL
1412 .get_acl = ll_get_acl,