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 *dir, struct dentry *dentry,
55 struct lookup_intent *it);
57 /* called from iget5_locked->find_inode() under inode_lock spinlock */
58 static int ll_test_inode(struct inode *inode, void *opaque)
60 struct ll_inode_info *lli = ll_i2info(inode);
61 struct lustre_md *md = opaque;
63 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
64 CERROR("MDS body missing FID\n");
68 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
74 static int ll_set_inode(struct inode *inode, void *opaque)
76 struct ll_inode_info *lli = ll_i2info(inode);
77 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
79 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
80 CERROR("MDS body missing FID\n");
84 lli->lli_fid = body->mbo_fid1;
85 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
86 CERROR("Can not initialize inode "DFID" without object type: "
88 PFID(&lli->lli_fid), body->mbo_valid);
92 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
93 if (unlikely(inode->i_mode == 0)) {
94 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
105 * Get an inode by inode number(@hash), which is already instantiated by
106 * the intent lookup).
108 struct inode *ll_iget(struct super_block *sb, ino_t hash,
109 struct lustre_md *md)
117 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
119 RETURN(ERR_PTR(-ENOMEM));
121 if (inode->i_state & I_NEW) {
122 rc = ll_read_inode2(inode, md);
123 if (rc == 0 && S_ISREG(inode->i_mode) &&
124 ll_i2info(inode)->lli_clob == NULL) {
125 CDEBUG(D_INODE, "%s: apply lsm %p to inode "DFID"\n",
126 ll_get_fsname(sb, NULL, 0), md->lsm,
127 PFID(ll_inode2fid(inode)));
128 rc = cl_file_inode_init(inode, md);
131 make_bad_inode(inode);
132 unlock_new_inode(inode);
136 unlock_new_inode(inode);
138 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
139 rc = ll_update_inode(inode, md);
140 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
141 PFID(&md->body->mbo_fid1), inode, rc);
143 make_bad_inode(inode);
152 static void ll_invalidate_negative_children(struct inode *dir)
154 struct dentry *dentry, *tmp_subdir;
155 DECLARE_LL_D_HLIST_NODE_PTR(p);
158 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
159 spin_lock(&dentry->d_lock);
160 if (!list_empty(&dentry->d_subdirs)) {
161 struct dentry *child;
163 list_for_each_entry_safe(child, tmp_subdir,
166 if (child->d_inode == NULL)
167 d_lustre_invalidate(child, 1);
170 spin_unlock(&dentry->d_lock);
172 ll_unlock_dcache(dir);
175 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
177 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
180 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
181 void *data, int flag)
183 struct lustre_handle lockh;
188 case LDLM_CB_BLOCKING:
189 ldlm_lock2handle(lock, &lockh);
190 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
192 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
196 case LDLM_CB_CANCELING: {
197 struct inode *inode = ll_inode_from_resource_lock(lock);
198 __u64 bits = lock->l_policy_data.l_inodebits.bits;
200 /* Inode is set to lock->l_resource->lr_lvb_inode
201 * for mdc - bug 24555 */
202 LASSERT(lock->l_ast_data == NULL);
207 /* Invalidate all dentries associated with this inode */
208 LASSERT(ldlm_is_canceling(lock));
210 if (!fid_res_name_eq(ll_inode2fid(inode),
211 &lock->l_resource->lr_name)) {
212 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
213 PFID(ll_inode2fid(inode)), inode);
217 if (bits & MDS_INODELOCK_XATTR) {
218 ll_xattr_cache_destroy(inode);
219 bits &= ~MDS_INODELOCK_XATTR;
222 /* For OPEN locks we differentiate between lock modes
223 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
224 if (bits & MDS_INODELOCK_OPEN)
225 ll_have_md_lock(inode, &bits, lock->l_req_mode);
227 if (bits & MDS_INODELOCK_OPEN) {
230 switch (lock->l_req_mode) {
241 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
245 ll_md_real_close(inode, fmode);
247 bits &= ~MDS_INODELOCK_OPEN;
250 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
251 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
252 ll_have_md_lock(inode, &bits, LCK_MINMODE);
254 if (bits & MDS_INODELOCK_LAYOUT) {
255 struct cl_object_conf conf = {
256 .coc_opc = OBJECT_CONF_INVALIDATE,
260 rc = ll_layout_conf(inode, &conf);
262 CDEBUG(D_INODE, "cannot invalidate layout of "
264 PFID(ll_inode2fid(inode)), rc);
267 if (bits & MDS_INODELOCK_UPDATE) {
268 struct ll_inode_info *lli = ll_i2info(inode);
270 spin_lock(&lli->lli_lock);
271 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
272 spin_unlock(&lli->lli_lock);
275 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
276 struct ll_inode_info *lli = ll_i2info(inode);
278 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
279 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
280 lli, PFID(&lli->lli_pfid));
281 truncate_inode_pages(inode->i_mapping, 0);
283 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
284 struct inode *master_inode = NULL;
287 /* This is slave inode, since all of the child
288 * dentry is connected on the master inode, so
289 * we have to invalidate the negative children
291 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
292 PFID(ll_inode2fid(inode)),
293 PFID(&lli->lli_pfid));
295 hash = cl_fid_build_ino(&lli->lli_pfid,
296 ll_need_32bit_api(ll_i2sbi(inode)));
298 master_inode = ilookup5(inode->i_sb, hash,
299 ll_test_inode_by_fid,
300 (void *)&lli->lli_pfid);
301 if (master_inode != NULL &&
302 !IS_ERR(master_inode)) {
303 ll_invalidate_negative_children(
308 ll_invalidate_negative_children(inode);
312 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
313 inode->i_sb->s_root != NULL &&
314 inode != inode->i_sb->s_root->d_inode)
315 ll_invalidate_aliases(inode);
327 __u32 ll_i2suppgid(struct inode *i)
329 if (in_group_p(i->i_gid))
330 return (__u32)from_kgid(&init_user_ns, i->i_gid);
332 return (__u32) __kgid_val(INVALID_GID);
335 /* Pack the required supplementary groups into the supplied groups array.
336 * If we don't need to use the groups from the target inode(s) then we
337 * instead pack one or more groups from the user's supplementary group
338 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
339 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
342 LASSERT(suppgids != NULL);
344 suppgids[0] = ll_i2suppgid(i1);
347 suppgids[1] = ll_i2suppgid(i2);
353 * try to reuse three types of dentry:
354 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
355 * by concurrent .revalidate).
356 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
357 * be cleared by others calling d_lustre_revalidate).
358 * 3. DISCONNECTED alias.
360 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
362 struct dentry *alias, *discon_alias, *invalid_alias;
363 DECLARE_LL_D_HLIST_NODE_PTR(p);
365 if (ll_d_hlist_empty(&inode->i_dentry))
368 discon_alias = invalid_alias = NULL;
370 ll_lock_dcache(inode);
371 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
372 LASSERT(alias != dentry);
374 spin_lock(&alias->d_lock);
375 if (alias->d_flags & DCACHE_DISCONNECTED)
376 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
377 discon_alias = alias;
378 else if (alias->d_parent == dentry->d_parent &&
379 alias->d_name.hash == dentry->d_name.hash &&
380 alias->d_name.len == dentry->d_name.len &&
381 memcmp(alias->d_name.name, dentry->d_name.name,
382 dentry->d_name.len) == 0)
383 invalid_alias = alias;
384 spin_unlock(&alias->d_lock);
389 alias = invalid_alias ?: discon_alias ?: NULL;
391 spin_lock(&alias->d_lock);
393 spin_unlock(&alias->d_lock);
395 ll_unlock_dcache(inode);
401 * Similar to d_splice_alias(), but lustre treats invalid alias
402 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
404 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
410 new = ll_find_alias(inode, de);
420 "Reuse dentry %p inode %p refc %d flags %#x\n",
421 new, new->d_inode, ll_d_count(new), new->d_flags);
429 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
430 de, de->d_inode, ll_d_count(de), de->d_flags);
434 static int ll_lookup_it_finish(struct ptlrpc_request *request,
435 struct lookup_intent *it,
436 struct inode *parent, struct dentry **de)
438 struct inode *inode = NULL;
443 /* NB 1 request reference will be taken away by ll_intent_lock()
445 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
446 it->d.lustre.it_disposition);
447 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
448 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
452 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
454 /* We used to query real size from OSTs here, but actually
455 this is not needed. For stat() calls size would be updated
456 from subsequent do_revalidate()->ll_inode_revalidate_it() in
458 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
459 Everybody else who needs correct file size would call
460 ll_glimpse_size or some equivalent themselves anyway.
461 Also see bug 7198. */
464 /* Only hash *de if it is unhashed (new dentry).
465 * Atoimc_open may passin hashed dentries for open.
467 if (d_unhashed(*de)) {
468 struct dentry *alias;
470 alias = ll_splice_alias(inode, *de);
472 RETURN(PTR_ERR(alias));
474 } else if (!it_disposition(it, DISP_LOOKUP_NEG) &&
475 !it_disposition(it, DISP_OPEN_CREATE)) {
476 /* With DISP_OPEN_CREATE dentry will
477 instantiated in ll_create_it. */
478 LASSERT((*de)->d_inode == NULL);
479 d_instantiate(*de, inode);
482 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
483 /* we have lookup look - unhide dentry */
484 if (bits & MDS_INODELOCK_LOOKUP)
485 d_lustre_revalidate(*de);
486 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
487 /* If file created on server, don't depend on parent UPDATE
488 * lock to unhide it. It is left hidden and next lookup can
489 * find it in ll_splice_alias.
491 /* Check that parent has UPDATE lock. */
492 struct lookup_intent parent_it = {
494 .d.lustre.it_lock_handle = 0 };
495 struct lu_fid fid = ll_i2info(parent)->lli_fid;
497 /* If it is striped directory, get the real stripe parent */
498 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
499 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
500 ll_i2info(parent)->lli_lsm_md,
502 (*de)->d_name.len, &fid);
507 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
509 d_lustre_revalidate(*de);
510 ll_intent_release(&parent_it);
517 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
518 struct lookup_intent *it)
520 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
521 struct dentry *save = dentry, *retval;
522 struct ptlrpc_request *req = NULL;
523 struct md_op_data *op_data;
528 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
529 RETURN(ERR_PTR(-ENAMETOOLONG));
531 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
532 dentry->d_name.len, dentry->d_name.name,
533 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
535 if (d_mountpoint(dentry))
536 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
538 if (it == NULL || it->it_op == IT_GETXATTR)
541 if (it->it_op == IT_GETATTR) {
542 rc = ll_statahead_enter(parent, &dentry, 0);
545 GOTO(out, retval = NULL);
546 GOTO(out, retval = dentry);
550 if (it->it_op & IT_CREAT)
551 opc = LUSTRE_OPC_CREATE;
553 opc = LUSTRE_OPC_ANY;
555 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
556 dentry->d_name.len, 0, opc, NULL);
558 RETURN((void *)op_data);
560 /* enforce umask if acl disabled or MDS doesn't support umask */
561 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
562 it->it_create_mode &= ~current_umask();
564 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
565 &ll_md_blocking_ast, 0);
566 ll_finish_md_op_data(op_data);
568 GOTO(out, retval = ERR_PTR(rc));
570 rc = ll_lookup_it_finish(req, it, parent, &dentry);
572 ll_intent_release(it);
573 GOTO(out, retval = ERR_PTR(rc));
576 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
577 !S_ISREG(dentry->d_inode->i_mode) &&
578 !S_ISDIR(dentry->d_inode->i_mode)) {
579 ll_release_openhandle(dentry, it);
581 ll_lookup_finish_locks(it, dentry);
584 GOTO(out, retval = NULL);
586 GOTO(out, retval = dentry);
589 ptlrpc_req_finished(req);
590 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
591 ll_statahead_mark(parent, dentry);
595 #ifdef HAVE_IOP_ATOMIC_OPEN
596 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
599 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
602 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
603 dentry->d_name.len, dentry->d_name.name,
604 PFID(ll_inode2fid(parent)), parent, flags);
606 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
607 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
610 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
614 de = ll_lookup_it(parent, dentry, itp);
617 ll_intent_release(itp);
623 * For cached negative dentry and new dentry, handle lookup/create/open
626 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
627 struct file *file, unsigned open_flags,
628 umode_t mode, int *opened)
630 struct lookup_intent *it;
632 long long lookup_flags = LOOKUP_OPEN;
636 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
637 "open_flags %x, mode %x opened %d\n",
638 dentry->d_name.len, dentry->d_name.name,
639 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
641 OBD_ALLOC(it, sizeof(*it));
646 if (open_flags & O_CREAT) {
647 it->it_op |= IT_CREAT;
648 lookup_flags |= LOOKUP_CREATE;
650 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
651 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
652 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
654 /* Dentry added to dcache tree in ll_lookup_it */
655 de = ll_lookup_it(dir, dentry, it);
662 if (it_disposition(it, DISP_OPEN_CREATE)) {
663 /* Dentry instantiated in ll_create_it. */
664 rc = ll_create_it(dir, dentry, it);
666 /* We dget in ll_splice_alias. */
672 *opened |= FILE_CREATED;
674 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
676 if (S_ISFIFO(dentry->d_inode->i_mode)) {
677 /* We cannot call open here as it would
680 if (it_disposition(it, DISP_ENQ_OPEN_REF))
682 (struct ptlrpc_request *)
683 it->d.lustre.it_data);
684 rc = finish_no_open(file, de);
686 file->private_data = it;
687 rc = finish_open(file, dentry, NULL, opened);
688 /* We dget in ll_splice_alias. finish_open takes
689 * care of dget for fd open.
695 rc = finish_no_open(file, de);
700 ll_intent_release(it);
701 OBD_FREE(it, sizeof(*it));
706 #else /* !HAVE_IOP_ATOMIC_OPEN */
707 static struct lookup_intent *
708 ll_convert_intent(struct open_intent *oit, int lookup_flags)
710 struct lookup_intent *it;
714 return ERR_PTR(-ENOMEM);
716 if (lookup_flags & LOOKUP_OPEN) {
718 if (lookup_flags & LOOKUP_CREATE)
719 it->it_op |= IT_CREAT;
720 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
721 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
722 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
724 it->it_op = IT_GETATTR;
730 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
731 struct nameidata *nd)
736 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
737 struct lookup_intent *it;
739 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
740 it = ll_d2d(dentry)->lld_it;
741 ll_d2d(dentry)->lld_it = NULL;
743 if ((nd->flags & LOOKUP_CREATE) &&
744 !(nd->flags & LOOKUP_OPEN))
747 it = ll_convert_intent(&nd->intent.open, nd->flags);
749 RETURN((struct dentry *)it);
752 de = ll_lookup_it(parent, dentry, it);
755 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
756 if (dentry->d_inode &&
757 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
758 if (S_ISFIFO(dentry->d_inode->i_mode)) {
759 // We cannot call open here as it would
762 (struct ptlrpc_request *)
763 it->d.lustre.it_data);
767 nd->intent.open.file->private_data = it;
768 filp = lookup_instantiate_filp(nd,
774 de = (struct dentry *)filp;
777 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
778 // XXX This can only reliably work on assumption
779 // that there are NO hashed negative dentries.
780 ll_d2d(dentry)->lld_it = it;
781 it = NULL; /* Will be freed in ll_create_nd */
782 /* We absolutely depend on ll_create_nd to be
783 * called to not leak this intent and possible
784 * data attached to it */
789 ll_intent_release(it);
790 OBD_FREE(it, sizeof(*it));
793 de = ll_lookup_it(parent, dentry, NULL);
798 #endif /* HAVE_IOP_ATOMIC_OPEN */
800 /* We depend on "mode" being set with the proper file type/umask by now */
801 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
803 struct inode *inode = NULL;
804 struct ptlrpc_request *request = NULL;
805 struct ll_sb_info *sbi = ll_i2sbi(dir);
809 LASSERT(it && it->d.lustre.it_disposition);
811 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
812 request = it->d.lustre.it_data;
813 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
814 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
816 GOTO(out, inode = ERR_PTR(rc));
818 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
820 /* We asked for a lock on the directory, but were granted a
821 * lock on the inode. Since we finally have an inode pointer,
822 * stuff it in the lock. */
823 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
824 PFID(ll_inode2fid(inode)), inode);
825 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
828 ptlrpc_req_finished(request);
833 * By the time this is called, we already have created the directory cache
834 * entry for the new file, but it is so far negative - it has no inode.
836 * We defer creating the OBD object(s) until open, to keep the intent and
837 * non-intent code paths similar, and also because we do not have the MDS
838 * inode number before calling ll_create_node() (which is needed for LOV),
839 * so we would need to do yet another RPC to the MDS to store the LOV EA
840 * data on the MDS. If needed, we would pass the PACKED lmm as data and
841 * lmm_size in datalen (the MDS still has code which will handle that).
843 * If the create succeeds, we fill in the inode information
844 * with d_instantiate().
846 static int ll_create_it(struct inode *dir, struct dentry *dentry,
847 struct lookup_intent *it)
853 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
854 dentry->d_name.len, dentry->d_name.name,
855 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
857 rc = it_open_error(DISP_OPEN_CREATE, it);
861 inode = ll_create_node(dir, it);
863 RETURN(PTR_ERR(inode));
865 d_instantiate(dentry, inode);
869 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
871 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
875 if (body->mbo_valid & OBD_MD_FLMTIME &&
876 body->mbo_mtime > LTIME_S(inode->i_mtime)) {
877 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
878 "\n", PFID(ll_inode2fid(inode)),
879 LTIME_S(inode->i_mtime), body->mbo_mtime);
880 LTIME_S(inode->i_mtime) = body->mbo_mtime;
883 if (body->mbo_valid & OBD_MD_FLCTIME &&
884 body->mbo_ctime > LTIME_S(inode->i_ctime))
885 LTIME_S(inode->i_ctime) = body->mbo_ctime;
888 static int ll_new_node(struct inode *dir, struct dentry *dchild,
889 const char *tgt, umode_t mode, int rdev, __u32 opc)
891 struct qstr *name = &dchild->d_name;
892 struct ptlrpc_request *request = NULL;
893 struct md_op_data *op_data;
894 struct inode *inode = NULL;
895 struct ll_sb_info *sbi = ll_i2sbi(dir);
900 if (unlikely(tgt != NULL))
901 tgt_len = strlen(tgt) + 1;
903 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
904 name->len, 0, opc, NULL);
906 GOTO(err_exit, err = PTR_ERR(op_data));
908 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
909 from_kuid(&init_user_ns, current_fsuid()),
910 from_kgid(&init_user_ns, current_fsgid()),
911 cfs_curproc_cap_pack(), rdev, &request);
912 ll_finish_md_op_data(op_data);
916 ll_update_times(request, dir);
918 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
922 d_instantiate(dchild, inode);
926 ptlrpc_req_finished(request);
931 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
934 struct qstr *name = &dchild->d_name;
938 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
939 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
942 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
943 mode &= ~current_umask();
945 switch (mode & S_IFMT) {
947 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
953 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
964 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
969 #ifdef HAVE_IOP_ATOMIC_OPEN
971 * Plain create. Intent create is handled in atomic_open.
973 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
974 umode_t mode, bool want_excl)
978 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
979 "flags=%u, excl=%d\n", dentry->d_name.len,
980 dentry->d_name.name, PFID(ll_inode2fid(dir)),
981 dir, mode, want_excl);
983 rc = ll_mknod(dir, dentry, mode, 0);
985 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
987 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
988 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
992 #else /* !HAVE_IOP_ATOMIC_OPEN */
993 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
994 ll_umode_t mode, struct nameidata *nd)
996 struct ll_dentry_data *lld = ll_d2d(dentry);
997 struct lookup_intent *it = NULL;
1004 return ll_mknod(dir, dentry, mode, 0);
1008 /* Was there an error? Propagate it! */
1009 if (it->d.lustre.it_status) {
1010 rc = it->d.lustre.it_status;
1014 rc = ll_create_it(dir, dentry, it);
1015 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1018 nd->intent.open.file->private_data = it;
1019 filp = lookup_instantiate_filp(nd, dentry, NULL);
1025 ll_intent_release(it);
1026 OBD_FREE(it, sizeof(*it));
1029 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1033 #endif /* HAVE_IOP_ATOMIC_OPEN */
1035 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1036 const char *oldpath)
1038 struct qstr *name = &dchild->d_name;
1042 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1043 name->len, name->name, PFID(ll_inode2fid(dir)),
1044 dir, 3000, oldpath);
1046 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1047 LUSTRE_OPC_SYMLINK);
1050 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1055 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1056 struct dentry *new_dentry)
1058 struct inode *src = old_dentry->d_inode;
1059 struct qstr *name = &new_dentry->d_name;
1060 struct ll_sb_info *sbi = ll_i2sbi(dir);
1061 struct ptlrpc_request *request = NULL;
1062 struct md_op_data *op_data;
1066 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1067 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1068 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1070 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1071 0, LUSTRE_OPC_ANY, NULL);
1072 if (IS_ERR(op_data))
1073 RETURN(PTR_ERR(op_data));
1075 err = md_link(sbi->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(sbi, LPROC_LL_LINK, 1);
1084 ptlrpc_req_finished(request);
1088 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1090 struct qstr *name = &dchild->d_name;
1094 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1095 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1097 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1098 mode &= ~current_umask();
1100 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1102 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1104 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1109 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1111 struct qstr *name = &dchild->d_name;
1112 struct ptlrpc_request *request = NULL;
1113 struct md_op_data *op_data;
1117 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1118 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1120 if (unlikely(d_mountpoint(dchild)))
1123 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1124 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1125 if (IS_ERR(op_data))
1126 RETURN(PTR_ERR(op_data));
1128 if (dchild->d_inode != NULL)
1129 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1131 op_data->op_fid2 = op_data->op_fid3;
1132 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1133 ll_finish_md_op_data(op_data);
1135 ll_update_times(request, dir);
1136 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1139 ptlrpc_req_finished(request);
1146 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1148 struct ptlrpc_request *request = NULL;
1149 struct md_op_data *op_data;
1153 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1154 namelen, name, PFID(ll_inode2fid(dir)), dir);
1156 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1157 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1158 if (IS_ERR(op_data))
1159 RETURN(PTR_ERR(op_data));
1160 op_data->op_cli_flags |= CLI_RM_ENTRY;
1161 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1162 ll_finish_md_op_data(op_data);
1164 ll_update_times(request, dir);
1165 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1168 ptlrpc_req_finished(request);
1172 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1174 struct mdt_body *body;
1175 struct lov_mds_md *eadata;
1176 struct lov_stripe_md *lsm = NULL;
1177 struct obd_trans_info oti = { 0 };
1179 struct obd_capa *oc = NULL;
1183 /* req is swabbed so this is safe */
1184 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1185 if (!(body->mbo_valid & OBD_MD_FLEASIZE))
1188 if (body->mbo_eadatasize == 0) {
1189 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1190 GOTO(out, rc = -EPROTO);
1193 /* The MDS sent back the EA because we unlinked the last reference
1194 * to this file. Use this EA to unlink the objects on the OST.
1195 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1196 * check it is complete and sensible. */
1197 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1198 body->mbo_eadatasize);
1199 LASSERT(eadata != NULL);
1201 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->mbo_eadatasize);
1203 CERROR("obd_unpackmd: %d\n", rc);
1206 LASSERT(rc >= sizeof(*lsm));
1210 GOTO(out_free_memmd, rc = -ENOMEM);
1212 oa->o_oi = lsm->lsm_oi;
1213 oa->o_mode = body->mbo_mode & S_IFMT;
1214 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1216 if (body->mbo_valid & OBD_MD_FLCOOKIE) {
1217 oa->o_valid |= OBD_MD_FLCOOKIE;
1218 oti.oti_logcookies =
1219 req_capsule_server_sized_get(&request->rq_pill,
1221 sizeof(struct llog_cookie) *
1222 lsm->lsm_stripe_count);
1223 if (oti.oti_logcookies == NULL) {
1224 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1225 body->mbo_valid &= ~OBD_MD_FLCOOKIE;
1229 if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
1230 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1232 GOTO(out_free_memmd, rc);
1235 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1236 ll_i2mdexp(dir), oc);
1239 CERROR("obd destroy objid "DOSTID" error %d\n",
1240 POSTID(&lsm->lsm_oi), rc);
1242 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1248 /* ll_unlink() doesn't update the inode with the new link count.
1249 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if
1250 * there is any lock existing. They will recycle dentries and inodes
1251 * based upon locks too. b=20433 */
1252 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1254 struct qstr *name = &dchild->d_name;
1255 struct ptlrpc_request *request = NULL;
1256 struct md_op_data *op_data;
1259 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1260 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1263 * XXX: unlink bind mountpoint maybe call to here,
1264 * just check it as vfs_unlink does.
1266 if (unlikely(d_mountpoint(dchild)))
1269 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1270 LUSTRE_OPC_ANY, NULL);
1271 if (IS_ERR(op_data))
1272 RETURN(PTR_ERR(op_data));
1274 if (dchild->d_inode != NULL)
1275 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1277 op_data->op_fid2 = op_data->op_fid3;
1278 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1279 ll_finish_md_op_data(op_data);
1283 ll_update_times(request, dir);
1284 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1286 rc = ll_objects_destroy(request, dir);
1288 ptlrpc_req_finished(request);
1292 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1293 struct inode *tgt, struct dentry *tgt_dchild)
1295 struct qstr *src_name = &src_dchild->d_name;
1296 struct qstr *tgt_name = &tgt_dchild->d_name;
1297 struct ptlrpc_request *request = NULL;
1298 struct ll_sb_info *sbi = ll_i2sbi(src);
1299 struct md_op_data *op_data;
1302 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1303 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1304 src_name->len, src_name->name,
1305 PFID(ll_inode2fid(src)), src, tgt_name->len,
1306 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1308 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1311 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1312 LUSTRE_OPC_ANY, NULL);
1313 if (IS_ERR(op_data))
1314 RETURN(PTR_ERR(op_data));
1316 if (src_dchild->d_inode != NULL)
1317 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1319 if (tgt_dchild->d_inode != NULL)
1320 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1322 err = md_rename(sbi->ll_md_exp, op_data,
1323 src_name->name, src_name->len,
1324 tgt_name->name, tgt_name->len, &request);
1325 ll_finish_md_op_data(op_data);
1327 ll_update_times(request, src);
1328 ll_update_times(request, tgt);
1329 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1330 err = ll_objects_destroy(request, src);
1333 ptlrpc_req_finished(request);
1336 d_move(src_dchild, tgt_dchild);
1341 const struct inode_operations ll_dir_inode_operations = {
1343 #ifdef HAVE_IOP_ATOMIC_OPEN
1344 .atomic_open = ll_atomic_open,
1346 .lookup = ll_lookup_nd,
1347 .create = ll_create_nd,
1348 /* We need all these non-raw things for NFSD, to not patch it. */
1349 .unlink = ll_unlink,
1352 .symlink = ll_symlink,
1354 .rename = ll_rename,
1355 .setattr = ll_setattr,
1356 .getattr = ll_getattr,
1357 .permission = ll_inode_permission,
1358 .setxattr = ll_setxattr,
1359 .getxattr = ll_getxattr,
1360 .listxattr = ll_listxattr,
1361 .removexattr = ll_removexattr,
1362 #ifdef HAVE_IOP_GET_ACL
1363 .get_acl = ll_get_acl,
1367 const struct inode_operations ll_special_inode_operations = {
1368 .setattr = ll_setattr,
1369 .getattr = ll_getattr,
1370 .permission = ll_inode_permission,
1371 .setxattr = ll_setxattr,
1372 .getxattr = ll_getxattr,
1373 .listxattr = ll_listxattr,
1374 .removexattr = ll_removexattr,
1375 #ifdef HAVE_IOP_GET_ACL
1376 .get_acl = ll_get_acl,