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, 2014, 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_dlm.h>
50 #include <lustre_ver.h>
51 #include "llite_internal.h"
53 static int ll_create_it(struct inode *dir, struct dentry *dentry,
54 struct lookup_intent *it);
56 /* called from iget5_locked->find_inode() under inode_lock spinlock */
57 static int ll_test_inode(struct inode *inode, void *opaque)
59 struct ll_inode_info *lli = ll_i2info(inode);
60 struct lustre_md *md = opaque;
62 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
63 CERROR("MDS body missing FID\n");
67 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
73 static int ll_set_inode(struct inode *inode, void *opaque)
75 struct ll_inode_info *lli = ll_i2info(inode);
76 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
78 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
79 CERROR("MDS body missing FID\n");
83 lli->lli_fid = body->mbo_fid1;
84 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
85 CERROR("Can not initialize inode "DFID" without object type: "
87 PFID(&lli->lli_fid), body->mbo_valid);
91 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
92 if (unlikely(inode->i_mode == 0)) {
93 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
104 * Get an inode by inode number(@hash), which is already instantiated by
105 * the intent lookup).
107 struct inode *ll_iget(struct super_block *sb, ino_t hash,
108 struct lustre_md *md)
116 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
118 RETURN(ERR_PTR(-ENOMEM));
120 if (inode->i_state & I_NEW) {
121 rc = ll_read_inode2(inode, md);
122 if (rc == 0 && S_ISREG(inode->i_mode) &&
123 ll_i2info(inode)->lli_clob == NULL) {
124 CDEBUG(D_INODE, "%s: apply lsm %p to inode "DFID"\n",
125 ll_get_fsname(sb, NULL, 0), md->lsm,
126 PFID(ll_inode2fid(inode)));
127 rc = cl_file_inode_init(inode, md);
130 make_bad_inode(inode);
131 unlock_new_inode(inode);
135 unlock_new_inode(inode);
137 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
138 rc = ll_update_inode(inode, md);
139 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
140 PFID(&md->body->mbo_fid1), inode, rc);
150 static void ll_invalidate_negative_children(struct inode *dir)
152 struct dentry *dentry, *tmp_subdir;
153 DECLARE_LL_D_HLIST_NODE_PTR(p);
156 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
157 spin_lock(&dentry->d_lock);
158 if (!list_empty(&dentry->d_subdirs)) {
159 struct dentry *child;
161 list_for_each_entry_safe(child, tmp_subdir,
164 if (child->d_inode == NULL)
165 d_lustre_invalidate(child, 1);
168 spin_unlock(&dentry->d_lock);
170 ll_unlock_dcache(dir);
173 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
175 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
178 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
179 void *data, int flag)
181 struct lustre_handle lockh;
186 case LDLM_CB_BLOCKING:
187 ldlm_lock2handle(lock, &lockh);
188 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
190 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
194 case LDLM_CB_CANCELING: {
195 struct inode *inode = ll_inode_from_resource_lock(lock);
196 __u64 bits = lock->l_policy_data.l_inodebits.bits;
198 /* Inode is set to lock->l_resource->lr_lvb_inode
199 * for mdc - bug 24555 */
200 LASSERT(lock->l_ast_data == NULL);
205 /* Invalidate all dentries associated with this inode */
206 LASSERT(ldlm_is_canceling(lock));
208 if (!fid_res_name_eq(ll_inode2fid(inode),
209 &lock->l_resource->lr_name)) {
210 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
211 PFID(ll_inode2fid(inode)), inode);
215 if (bits & MDS_INODELOCK_XATTR) {
216 if (S_ISDIR(inode->i_mode))
217 ll_i2info(inode)->lli_def_stripe_offset = -1;
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) && S_ISDIR(inode->i_mode)) {
268 struct ll_inode_info *lli = ll_i2info(inode);
270 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
271 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
272 lli, PFID(&lli->lli_pfid));
273 truncate_inode_pages(inode->i_mapping, 0);
275 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
276 struct inode *master_inode = NULL;
279 /* This is slave inode, since all of the child
280 * dentry is connected on the master inode, so
281 * we have to invalidate the negative children
283 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
284 PFID(ll_inode2fid(inode)),
285 PFID(&lli->lli_pfid));
287 hash = cl_fid_build_ino(&lli->lli_pfid,
288 ll_need_32bit_api(ll_i2sbi(inode)));
290 master_inode = ilookup5(inode->i_sb, hash,
291 ll_test_inode_by_fid,
292 (void *)&lli->lli_pfid);
293 if (master_inode != NULL &&
294 !IS_ERR(master_inode)) {
295 ll_invalidate_negative_children(
300 ll_invalidate_negative_children(inode);
304 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
305 inode->i_sb->s_root != NULL &&
306 inode != inode->i_sb->s_root->d_inode)
307 ll_invalidate_aliases(inode);
319 __u32 ll_i2suppgid(struct inode *i)
321 if (in_group_p(i->i_gid))
322 return (__u32)from_kgid(&init_user_ns, i->i_gid);
324 return (__u32) __kgid_val(INVALID_GID);
327 /* Pack the required supplementary groups into the supplied groups array.
328 * If we don't need to use the groups from the target inode(s) then we
329 * instead pack one or more groups from the user's supplementary group
330 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
331 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
334 LASSERT(suppgids != NULL);
336 suppgids[0] = ll_i2suppgid(i1);
339 suppgids[1] = ll_i2suppgid(i2);
345 * try to reuse three types of dentry:
346 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
347 * by concurrent .revalidate).
348 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
349 * be cleared by others calling d_lustre_revalidate).
350 * 3. DISCONNECTED alias.
352 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
354 struct dentry *alias, *discon_alias, *invalid_alias;
355 DECLARE_LL_D_HLIST_NODE_PTR(p);
357 if (ll_d_hlist_empty(&inode->i_dentry))
360 discon_alias = invalid_alias = NULL;
362 ll_lock_dcache(inode);
363 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
364 LASSERT(alias != dentry);
366 spin_lock(&alias->d_lock);
367 if (alias->d_flags & DCACHE_DISCONNECTED)
368 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
369 discon_alias = alias;
370 else if (alias->d_parent == dentry->d_parent &&
371 alias->d_name.hash == dentry->d_name.hash &&
372 alias->d_name.len == dentry->d_name.len &&
373 memcmp(alias->d_name.name, dentry->d_name.name,
374 dentry->d_name.len) == 0)
375 invalid_alias = alias;
376 spin_unlock(&alias->d_lock);
381 alias = invalid_alias ?: discon_alias ?: NULL;
383 spin_lock(&alias->d_lock);
385 spin_unlock(&alias->d_lock);
387 ll_unlock_dcache(inode);
393 * Similar to d_splice_alias(), but lustre treats invalid alias
394 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
396 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
402 new = ll_find_alias(inode, de);
412 "Reuse dentry %p inode %p refc %d flags %#x\n",
413 new, new->d_inode, ll_d_count(new), new->d_flags);
421 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
422 de, de->d_inode, ll_d_count(de), de->d_flags);
426 static int ll_lookup_it_finish(struct ptlrpc_request *request,
427 struct lookup_intent *it,
428 struct inode *parent, struct dentry **de)
430 struct inode *inode = NULL;
435 /* NB 1 request reference will be taken away by ll_intent_lock()
437 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
438 it->d.lustre.it_disposition);
439 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
440 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
444 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
446 /* We used to query real size from OSTs here, but actually
447 this is not needed. For stat() calls size would be updated
448 from subsequent do_revalidate()->ll_inode_revalidate_it() in
450 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
451 Everybody else who needs correct file size would call
452 ll_glimpse_size or some equivalent themselves anyway.
453 Also see bug 7198. */
456 /* Only hash *de if it is unhashed (new dentry).
457 * Atoimc_open may passin hashed dentries for open.
459 if (d_unhashed(*de)) {
460 struct dentry *alias;
462 alias = ll_splice_alias(inode, *de);
464 GOTO(out, rc = PTR_ERR(alias));
467 } else if (!it_disposition(it, DISP_LOOKUP_NEG) &&
468 !it_disposition(it, DISP_OPEN_CREATE)) {
469 /* With DISP_OPEN_CREATE dentry will
470 instantiated in ll_create_it. */
471 LASSERT((*de)->d_inode == NULL);
472 d_instantiate(*de, inode);
475 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
476 /* we have lookup look - unhide dentry */
477 if (bits & MDS_INODELOCK_LOOKUP)
478 d_lustre_revalidate(*de);
479 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
480 /* If file created on server, don't depend on parent UPDATE
481 * lock to unhide it. It is left hidden and next lookup can
482 * find it in ll_splice_alias.
484 /* Check that parent has UPDATE lock. */
485 struct lookup_intent parent_it = {
487 .d.lustre.it_lock_handle = 0 };
488 struct lu_fid fid = ll_i2info(parent)->lli_fid;
490 /* If it is striped directory, get the real stripe parent */
491 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
492 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
493 ll_i2info(parent)->lli_lsm_md,
495 (*de)->d_name.len, &fid);
500 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
502 d_lustre_revalidate(*de);
503 ll_intent_release(&parent_it);
510 if (rc != 0 && it->it_op & IT_OPEN)
511 ll_open_cleanup((*de)->d_sb, request);
516 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
517 struct lookup_intent *it)
519 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
520 struct dentry *save = dentry, *retval;
521 struct ptlrpc_request *req = NULL;
522 struct md_op_data *op_data = NULL;
527 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
528 RETURN(ERR_PTR(-ENAMETOOLONG));
530 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
531 dentry->d_name.len, dentry->d_name.name,
532 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
534 if (d_mountpoint(dentry))
535 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
537 if (it == NULL || it->it_op == IT_GETXATTR)
540 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
541 rc = ll_statahead(parent, &dentry, 0);
543 RETURN(dentry == save ? NULL : dentry);
546 if (it->it_op & IT_CREAT)
547 opc = LUSTRE_OPC_CREATE;
549 opc = LUSTRE_OPC_ANY;
551 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
552 dentry->d_name.len, 0, opc, NULL);
554 RETURN((void *)op_data);
556 /* enforce umask if acl disabled or MDS doesn't support umask */
557 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
558 it->it_create_mode &= ~current_umask();
560 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
561 &ll_md_blocking_ast, 0);
562 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
563 * client does not know which suppgid should be sent to the MDS, or
564 * some other(s) changed the target file's GID after this RPC sent
565 * to the MDS with the suppgid as the original GID, then we should
566 * try again with right suppgid. */
567 if (rc == -EACCES && it->it_op & IT_OPEN &&
568 it_disposition(it, DISP_OPEN_DENY)) {
569 struct mdt_body *body;
571 LASSERT(req != NULL);
573 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
574 if (op_data->op_suppgids[0] == body->mbo_gid ||
575 op_data->op_suppgids[1] == body->mbo_gid ||
576 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
577 GOTO(out, retval = ERR_PTR(-EACCES));
579 fid_zero(&op_data->op_fid2);
580 op_data->op_suppgids[1] = body->mbo_gid;
581 ptlrpc_req_finished(req);
583 ll_intent_release(it);
584 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
585 &ll_md_blocking_ast, 0);
589 GOTO(out, retval = ERR_PTR(rc));
591 rc = ll_lookup_it_finish(req, it, parent, &dentry);
593 ll_intent_release(it);
594 GOTO(out, retval = ERR_PTR(rc));
597 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
598 !S_ISREG(dentry->d_inode->i_mode) &&
599 !S_ISDIR(dentry->d_inode->i_mode)) {
600 ll_release_openhandle(dentry, it);
602 ll_lookup_finish_locks(it, dentry);
604 GOTO(out, retval = (dentry == save) ? NULL : dentry);
607 if (op_data != NULL && !IS_ERR(op_data))
608 ll_finish_md_op_data(op_data);
610 ptlrpc_req_finished(req);
614 #ifdef HAVE_IOP_ATOMIC_OPEN
615 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
618 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
621 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
622 dentry->d_name.len, dentry->d_name.name,
623 PFID(ll_inode2fid(parent)), parent, flags);
625 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
626 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
629 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
633 de = ll_lookup_it(parent, dentry, itp);
636 ll_intent_release(itp);
642 * For cached negative dentry and new dentry, handle lookup/create/open
645 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
646 struct file *file, unsigned open_flags,
647 umode_t mode, int *opened)
649 struct lookup_intent *it;
651 long long lookup_flags = LOOKUP_OPEN;
655 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
656 "open_flags %x, mode %x opened %d\n",
657 dentry->d_name.len, dentry->d_name.name,
658 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
660 OBD_ALLOC(it, sizeof(*it));
665 if (open_flags & O_CREAT) {
666 it->it_op |= IT_CREAT;
667 lookup_flags |= LOOKUP_CREATE;
669 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
670 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
671 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
673 /* Dentry added to dcache tree in ll_lookup_it */
674 de = ll_lookup_it(dir, dentry, it);
681 if (it_disposition(it, DISP_OPEN_CREATE)) {
682 /* Dentry instantiated in ll_create_it. */
683 rc = ll_create_it(dir, dentry, it);
685 /* We dget in ll_splice_alias. */
691 *opened |= FILE_CREATED;
693 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
695 if (S_ISFIFO(dentry->d_inode->i_mode)) {
696 /* We cannot call open here as it would
699 if (it_disposition(it, DISP_ENQ_OPEN_REF))
701 (struct ptlrpc_request *)
702 it->d.lustre.it_data);
703 rc = finish_no_open(file, de);
705 file->private_data = it;
706 rc = finish_open(file, dentry, NULL, opened);
707 /* We dget in ll_splice_alias. finish_open takes
708 * care of dget for fd open.
714 rc = finish_no_open(file, de);
719 ll_intent_release(it);
720 OBD_FREE(it, sizeof(*it));
725 #else /* !HAVE_IOP_ATOMIC_OPEN */
726 static struct lookup_intent *
727 ll_convert_intent(struct open_intent *oit, int lookup_flags)
729 struct lookup_intent *it;
733 return ERR_PTR(-ENOMEM);
735 if (lookup_flags & LOOKUP_OPEN) {
737 if (lookup_flags & LOOKUP_CREATE)
738 it->it_op |= IT_CREAT;
739 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
740 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
741 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
743 it->it_op = IT_GETATTR;
749 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
750 struct nameidata *nd)
755 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
756 struct lookup_intent *it;
758 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
759 it = ll_d2d(dentry)->lld_it;
760 ll_d2d(dentry)->lld_it = NULL;
762 if ((nd->flags & LOOKUP_CREATE) &&
763 !(nd->flags & LOOKUP_OPEN))
766 it = ll_convert_intent(&nd->intent.open, nd->flags);
768 RETURN((struct dentry *)it);
771 de = ll_lookup_it(parent, dentry, it);
774 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
775 if (dentry->d_inode &&
776 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
777 if (S_ISFIFO(dentry->d_inode->i_mode)) {
778 // We cannot call open here as it would
781 (struct ptlrpc_request *)
782 it->d.lustre.it_data);
786 nd->intent.open.file->private_data = it;
787 filp = lookup_instantiate_filp(nd,
793 de = (struct dentry *)filp;
796 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
797 // XXX This can only reliably work on assumption
798 // that there are NO hashed negative dentries.
799 ll_d2d(dentry)->lld_it = it;
800 it = NULL; /* Will be freed in ll_create_nd */
801 /* We absolutely depend on ll_create_nd to be
802 * called to not leak this intent and possible
803 * data attached to it */
808 ll_intent_release(it);
809 OBD_FREE(it, sizeof(*it));
812 de = ll_lookup_it(parent, dentry, NULL);
817 #endif /* HAVE_IOP_ATOMIC_OPEN */
819 /* We depend on "mode" being set with the proper file type/umask by now */
820 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
822 struct inode *inode = NULL;
823 struct ptlrpc_request *request = NULL;
824 struct ll_sb_info *sbi = ll_i2sbi(dir);
828 LASSERT(it && it->d.lustre.it_disposition);
830 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
831 request = it->d.lustre.it_data;
832 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
833 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
835 GOTO(out, inode = ERR_PTR(rc));
837 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
839 /* We asked for a lock on the directory, but were granted a
840 * lock on the inode. Since we finally have an inode pointer,
841 * stuff it in the lock. */
842 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
843 PFID(ll_inode2fid(inode)), inode);
844 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
847 ptlrpc_req_finished(request);
852 * By the time this is called, we already have created the directory cache
853 * entry for the new file, but it is so far negative - it has no inode.
855 * We defer creating the OBD object(s) until open, to keep the intent and
856 * non-intent code paths similar, and also because we do not have the MDS
857 * inode number before calling ll_create_node() (which is needed for LOV),
858 * so we would need to do yet another RPC to the MDS to store the LOV EA
859 * data on the MDS. If needed, we would pass the PACKED lmm as data and
860 * lmm_size in datalen (the MDS still has code which will handle that).
862 * If the create succeeds, we fill in the inode information
863 * with d_instantiate().
865 static int ll_create_it(struct inode *dir, struct dentry *dentry,
866 struct lookup_intent *it)
872 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
873 dentry->d_name.len, dentry->d_name.name,
874 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
876 rc = it_open_error(DISP_OPEN_CREATE, it);
880 inode = ll_create_node(dir, it);
882 RETURN(PTR_ERR(inode));
884 d_instantiate(dentry, inode);
888 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
890 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
894 if (body->mbo_valid & OBD_MD_FLMTIME &&
895 body->mbo_mtime > LTIME_S(inode->i_mtime)) {
896 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
897 "\n", PFID(ll_inode2fid(inode)),
898 LTIME_S(inode->i_mtime), body->mbo_mtime);
899 LTIME_S(inode->i_mtime) = body->mbo_mtime;
902 if (body->mbo_valid & OBD_MD_FLCTIME &&
903 body->mbo_ctime > LTIME_S(inode->i_ctime))
904 LTIME_S(inode->i_ctime) = body->mbo_ctime;
907 static int ll_new_node(struct inode *dir, struct dentry *dchild,
908 const char *tgt, umode_t mode, int rdev, __u32 opc)
910 struct qstr *name = &dchild->d_name;
911 struct ptlrpc_request *request = NULL;
912 struct md_op_data *op_data;
913 struct inode *inode = NULL;
914 struct ll_sb_info *sbi = ll_i2sbi(dir);
919 if (unlikely(tgt != NULL))
920 tgt_len = strlen(tgt) + 1;
923 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
924 name->len, 0, opc, NULL);
926 GOTO(err_exit, err = PTR_ERR(op_data));
928 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
929 from_kuid(&init_user_ns, current_fsuid()),
930 from_kgid(&init_user_ns, current_fsgid()),
931 cfs_curproc_cap_pack(), rdev, &request);
932 ll_finish_md_op_data(op_data);
934 /* If the client doesn't know where to create a subdirectory (or
935 * in case of a race that sends the RPC to the wrong MDS), the
936 * MDS will return -EREMOTE and the client will fetch the layout
937 * for the directory, either from the local xattr cache or the
938 * MDS, then create the directory on the right MDT. */
939 if (err == -EREMOTE) {
940 struct lmv_user_md *lum;
943 ptlrpc_req_finished(request);
948 GOTO(err_exit, err = -ENOMEM);
950 rc = ll_getxattr_common(dir, XATTR_NAME_DEFAULT_LMV,
961 ll_i2info(dir)->lli_def_stripe_offset =
962 le32_to_cpu(lum->lum_stripe_offset);
969 ll_update_times(request, dir);
971 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
975 d_instantiate(dchild, inode);
980 ptlrpc_req_finished(request);
985 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
988 struct qstr *name = &dchild->d_name;
992 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
993 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
996 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
997 mode &= ~current_umask();
999 switch (mode & S_IFMT) {
1001 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1007 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1018 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1023 #ifdef HAVE_IOP_ATOMIC_OPEN
1025 * Plain create. Intent create is handled in atomic_open.
1027 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1028 umode_t mode, bool want_excl)
1032 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1033 "flags=%u, excl=%d\n", dentry->d_name.len,
1034 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1035 dir, mode, want_excl);
1037 rc = ll_mknod(dir, dentry, mode, 0);
1039 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1041 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1042 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1046 #else /* !HAVE_IOP_ATOMIC_OPEN */
1047 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1048 ll_umode_t mode, struct nameidata *nd)
1050 struct ll_dentry_data *lld = ll_d2d(dentry);
1051 struct lookup_intent *it = NULL;
1058 return ll_mknod(dir, dentry, mode, 0);
1062 /* Was there an error? Propagate it! */
1063 if (it->d.lustre.it_status) {
1064 rc = it->d.lustre.it_status;
1068 rc = ll_create_it(dir, dentry, it);
1069 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1072 nd->intent.open.file->private_data = it;
1073 filp = lookup_instantiate_filp(nd, dentry, NULL);
1079 ll_intent_release(it);
1080 OBD_FREE(it, sizeof(*it));
1083 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1087 #endif /* HAVE_IOP_ATOMIC_OPEN */
1089 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1090 const char *oldpath)
1092 struct qstr *name = &dchild->d_name;
1096 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1097 name->len, name->name, PFID(ll_inode2fid(dir)),
1098 dir, 3000, oldpath);
1100 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1101 LUSTRE_OPC_SYMLINK);
1104 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1109 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1110 struct dentry *new_dentry)
1112 struct inode *src = old_dentry->d_inode;
1113 struct qstr *name = &new_dentry->d_name;
1114 struct ll_sb_info *sbi = ll_i2sbi(dir);
1115 struct ptlrpc_request *request = NULL;
1116 struct md_op_data *op_data;
1120 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1121 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1122 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1124 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1125 0, LUSTRE_OPC_ANY, NULL);
1126 if (IS_ERR(op_data))
1127 RETURN(PTR_ERR(op_data));
1129 err = md_link(sbi->ll_md_exp, op_data, &request);
1130 ll_finish_md_op_data(op_data);
1134 ll_update_times(request, dir);
1135 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1138 ptlrpc_req_finished(request);
1142 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1144 struct qstr *name = &dchild->d_name;
1148 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1149 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1151 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1152 mode &= ~current_umask();
1154 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1156 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1158 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1163 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1165 struct qstr *name = &dchild->d_name;
1166 struct ptlrpc_request *request = NULL;
1167 struct md_op_data *op_data;
1171 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1172 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1174 if (unlikely(d_mountpoint(dchild)))
1177 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1178 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1179 if (IS_ERR(op_data))
1180 RETURN(PTR_ERR(op_data));
1182 if (dchild->d_inode != NULL)
1183 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1185 op_data->op_fid2 = op_data->op_fid3;
1186 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1187 ll_finish_md_op_data(op_data);
1189 ll_update_times(request, dir);
1190 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1193 ptlrpc_req_finished(request);
1200 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1202 struct ptlrpc_request *request = NULL;
1203 struct md_op_data *op_data;
1207 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1208 namelen, name, PFID(ll_inode2fid(dir)), dir);
1210 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1211 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1212 if (IS_ERR(op_data))
1213 RETURN(PTR_ERR(op_data));
1214 op_data->op_cli_flags |= CLI_RM_ENTRY;
1215 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1216 ll_finish_md_op_data(op_data);
1218 ll_update_times(request, dir);
1219 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1222 ptlrpc_req_finished(request);
1226 /* ll_unlink() doesn't update the inode with the new link count.
1227 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if
1228 * there is any lock existing. They will recycle dentries and inodes
1229 * based upon locks too. b=20433 */
1230 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1232 struct qstr *name = &dchild->d_name;
1233 struct ptlrpc_request *request = NULL;
1234 struct md_op_data *op_data;
1237 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1238 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1241 * XXX: unlink bind mountpoint maybe call to here,
1242 * just check it as vfs_unlink does.
1244 if (unlikely(d_mountpoint(dchild)))
1247 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1248 LUSTRE_OPC_ANY, NULL);
1249 if (IS_ERR(op_data))
1250 RETURN(PTR_ERR(op_data));
1252 if (dchild->d_inode != NULL)
1253 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1255 op_data->op_fid2 = op_data->op_fid3;
1256 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1257 ll_finish_md_op_data(op_data);
1261 ll_update_times(request, dir);
1262 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1265 ptlrpc_req_finished(request);
1269 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1270 struct inode *tgt, struct dentry *tgt_dchild)
1272 struct qstr *src_name = &src_dchild->d_name;
1273 struct qstr *tgt_name = &tgt_dchild->d_name;
1274 struct ptlrpc_request *request = NULL;
1275 struct ll_sb_info *sbi = ll_i2sbi(src);
1276 struct md_op_data *op_data;
1279 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1280 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1281 src_name->len, src_name->name,
1282 PFID(ll_inode2fid(src)), src, tgt_name->len,
1283 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1285 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1288 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1289 LUSTRE_OPC_ANY, NULL);
1290 if (IS_ERR(op_data))
1291 RETURN(PTR_ERR(op_data));
1293 if (src_dchild->d_inode != NULL)
1294 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1296 if (tgt_dchild->d_inode != NULL)
1297 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1299 err = md_rename(sbi->ll_md_exp, op_data,
1300 src_name->name, src_name->len,
1301 tgt_name->name, tgt_name->len, &request);
1302 ll_finish_md_op_data(op_data);
1304 ll_update_times(request, src);
1305 ll_update_times(request, tgt);
1306 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1309 ptlrpc_req_finished(request);
1312 d_move(src_dchild, tgt_dchild);
1317 const struct inode_operations ll_dir_inode_operations = {
1319 #ifdef HAVE_IOP_ATOMIC_OPEN
1320 .atomic_open = ll_atomic_open,
1322 .lookup = ll_lookup_nd,
1323 .create = ll_create_nd,
1324 /* We need all these non-raw things for NFSD, to not patch it. */
1325 .unlink = ll_unlink,
1328 .symlink = ll_symlink,
1330 .rename = ll_rename,
1331 .setattr = ll_setattr,
1332 .getattr = ll_getattr,
1333 .permission = ll_inode_permission,
1334 .setxattr = ll_setxattr,
1335 .getxattr = ll_getxattr,
1336 .listxattr = ll_listxattr,
1337 .removexattr = ll_removexattr,
1338 #ifdef HAVE_IOP_GET_ACL
1339 .get_acl = ll_get_acl,
1343 const struct inode_operations ll_special_inode_operations = {
1344 .setattr = ll_setattr,
1345 .getattr = ll_getattr,
1346 .permission = ll_inode_permission,
1347 .setxattr = ll_setxattr,
1348 .getxattr = ll_getxattr,
1349 .listxattr = ll_listxattr,
1350 .removexattr = ll_removexattr,
1351 #ifdef HAVE_IOP_GET_ACL
1352 .get_acl = ll_get_acl,