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>
44 #include <linux/user_namespace.h>
45 #ifdef HAVE_UIDGID_HEADER
46 # include <linux/uidgid.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include <obd_support.h>
52 #include <lustre_fid.h>
53 #include <lustre_dlm.h>
54 #include <lustre_ver.h>
55 #include "llite_internal.h"
57 static int ll_create_it(struct inode *dir, struct dentry *dentry,
58 struct lookup_intent *it);
60 /* called from iget5_locked->find_inode() under inode_lock spinlock */
61 static int ll_test_inode(struct inode *inode, void *opaque)
63 struct ll_inode_info *lli = ll_i2info(inode);
64 struct lustre_md *md = opaque;
66 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
67 CERROR("MDS body missing FID\n");
71 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
77 static int ll_set_inode(struct inode *inode, void *opaque)
79 struct ll_inode_info *lli = ll_i2info(inode);
80 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
82 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
83 CERROR("MDS body missing FID\n");
87 lli->lli_fid = body->mbo_fid1;
88 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
89 CERROR("Can not initialize inode "DFID" without object type: "
91 PFID(&lli->lli_fid), body->mbo_valid);
95 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
96 if (unlikely(inode->i_mode == 0)) {
97 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
108 * Get an inode by inode number(@hash), which is already instantiated by
109 * the intent lookup).
111 struct inode *ll_iget(struct super_block *sb, ino_t hash,
112 struct lustre_md *md)
120 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
122 RETURN(ERR_PTR(-ENOMEM));
124 if (inode->i_state & I_NEW) {
125 rc = ll_read_inode2(inode, md);
126 if (rc == 0 && S_ISREG(inode->i_mode) &&
127 ll_i2info(inode)->lli_clob == NULL)
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);
151 static void ll_invalidate_negative_children(struct inode *dir)
153 struct dentry *dentry, *tmp_subdir;
154 DECLARE_LL_D_HLIST_NODE_PTR(p);
157 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
158 spin_lock(&dentry->d_lock);
159 if (!list_empty(&dentry->d_subdirs)) {
160 struct dentry *child;
162 list_for_each_entry_safe(child, tmp_subdir,
165 if (child->d_inode == NULL)
166 d_lustre_invalidate(child, 1);
169 spin_unlock(&dentry->d_lock);
171 ll_unlock_dcache(dir);
174 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
176 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
179 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
180 void *data, int flag)
182 struct lustre_handle lockh;
187 case LDLM_CB_BLOCKING:
188 ldlm_lock2handle(lock, &lockh);
189 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
191 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
195 case LDLM_CB_CANCELING: {
196 struct inode *inode = ll_inode_from_resource_lock(lock);
197 __u64 bits = lock->l_policy_data.l_inodebits.bits;
199 /* Inode is set to lock->l_resource->lr_lvb_inode
200 * for mdc - bug 24555 */
201 LASSERT(lock->l_ast_data == NULL);
206 /* Invalidate all dentries associated with this inode */
207 LASSERT(ldlm_is_canceling(lock));
209 if (!fid_res_name_eq(ll_inode2fid(inode),
210 &lock->l_resource->lr_name)) {
211 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
212 PFID(ll_inode2fid(inode)), inode);
216 if (bits & MDS_INODELOCK_XATTR) {
217 if (S_ISDIR(inode->i_mode))
218 ll_i2info(inode)->lli_def_stripe_offset = -1;
219 ll_xattr_cache_destroy(inode);
220 bits &= ~MDS_INODELOCK_XATTR;
223 /* For OPEN locks we differentiate between lock modes
224 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
225 if (bits & MDS_INODELOCK_OPEN)
226 ll_have_md_lock(inode, &bits, lock->l_req_mode);
228 if (bits & MDS_INODELOCK_OPEN) {
231 switch (lock->l_req_mode) {
242 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
246 ll_md_real_close(inode, fmode);
248 bits &= ~MDS_INODELOCK_OPEN;
251 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
252 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
253 ll_have_md_lock(inode, &bits, LCK_MINMODE);
255 if (bits & MDS_INODELOCK_LAYOUT) {
256 struct cl_object_conf conf = {
257 .coc_opc = OBJECT_CONF_INVALIDATE,
261 rc = ll_layout_conf(inode, &conf);
263 CDEBUG(D_INODE, "cannot invalidate layout of "
265 PFID(ll_inode2fid(inode)), rc);
268 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
269 struct ll_inode_info *lli = ll_i2info(inode);
271 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
272 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
273 lli, PFID(&lli->lli_pfid));
274 truncate_inode_pages(inode->i_mapping, 0);
276 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
277 struct inode *master_inode = NULL;
280 /* This is slave inode, since all of the child
281 * dentry is connected on the master inode, so
282 * we have to invalidate the negative children
284 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
285 PFID(ll_inode2fid(inode)),
286 PFID(&lli->lli_pfid));
288 hash = cl_fid_build_ino(&lli->lli_pfid,
289 ll_need_32bit_api(ll_i2sbi(inode)));
291 /* Do not lookup the inode with ilookup5,
292 * otherwise it will cause dead lock,
294 * 1. Client1 send chmod req to the MDT0, then
295 * on MDT0, it enqueues master and all of its
296 * slaves lock, (mdt_attr_set() ->
297 * mdt_lock_slaves()), after gets master and
298 * stripe0 lock, it will send the enqueue req
299 * (for stripe1) to MDT1, then MDT1 finds the
300 * lock has been granted to client2. Then MDT1
301 * sends blocking ast to client2.
303 * 2. At the same time, client2 tries to unlink
304 * the striped dir (rm -rf striped_dir), and
305 * during lookup, it will hold the master inode
306 * of the striped directory, whose inode state
307 * is NEW, then tries to revalidate all of its
308 * slaves, (ll_prep_inode()->ll_iget()->
309 * ll_read_inode2()-> ll_update_inode().). And
310 * it will be blocked on the server side because
313 * 3. Then the client get the blocking_ast req,
314 * cancel the lock, but being blocked if using
315 * ->ilookup5()), because master inode state is
317 master_inode = ilookup5_nowait(inode->i_sb,
318 hash, ll_test_inode_by_fid,
319 (void *)&lli->lli_pfid);
320 if (master_inode != NULL &&
321 !IS_ERR(master_inode)) {
322 ll_invalidate_negative_children(
327 ll_invalidate_negative_children(inode);
331 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
332 inode->i_sb->s_root != NULL &&
333 inode != inode->i_sb->s_root->d_inode)
334 ll_invalidate_aliases(inode);
346 __u32 ll_i2suppgid(struct inode *i)
348 if (in_group_p(i->i_gid))
349 return (__u32)from_kgid(&init_user_ns, i->i_gid);
351 return (__u32) __kgid_val(INVALID_GID);
354 /* Pack the required supplementary groups into the supplied groups array.
355 * If we don't need to use the groups from the target inode(s) then we
356 * instead pack one or more groups from the user's supplementary group
357 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
358 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
361 LASSERT(suppgids != NULL);
363 suppgids[0] = ll_i2suppgid(i1);
366 suppgids[1] = ll_i2suppgid(i2);
372 * try to reuse three types of dentry:
373 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
374 * by concurrent .revalidate).
375 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
376 * be cleared by others calling d_lustre_revalidate).
377 * 3. DISCONNECTED alias.
379 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
381 struct dentry *alias, *discon_alias, *invalid_alias;
382 DECLARE_LL_D_HLIST_NODE_PTR(p);
384 if (ll_d_hlist_empty(&inode->i_dentry))
387 discon_alias = invalid_alias = NULL;
389 ll_lock_dcache(inode);
390 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
391 LASSERT(alias != dentry);
393 spin_lock(&alias->d_lock);
394 if (alias->d_flags & DCACHE_DISCONNECTED)
395 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
396 discon_alias = alias;
397 else if (alias->d_parent == dentry->d_parent &&
398 alias->d_name.hash == dentry->d_name.hash &&
399 alias->d_name.len == dentry->d_name.len &&
400 memcmp(alias->d_name.name, dentry->d_name.name,
401 dentry->d_name.len) == 0)
402 invalid_alias = alias;
403 spin_unlock(&alias->d_lock);
408 alias = invalid_alias ?: discon_alias ?: NULL;
410 spin_lock(&alias->d_lock);
412 spin_unlock(&alias->d_lock);
414 ll_unlock_dcache(inode);
420 * Similar to d_splice_alias(), but lustre treats invalid alias
421 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
423 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
429 new = ll_find_alias(inode, de);
439 "Reuse dentry %p inode %p refc %d flags %#x\n",
440 new, new->d_inode, ll_d_count(new), new->d_flags);
448 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
449 de, de->d_inode, ll_d_count(de), de->d_flags);
453 static int ll_lookup_it_finish(struct ptlrpc_request *request,
454 struct lookup_intent *it,
455 struct inode *parent, struct dentry **de)
457 struct inode *inode = NULL;
462 /* NB 1 request reference will be taken away by ll_intent_lock()
464 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
465 it->d.lustre.it_disposition);
466 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
467 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
471 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
473 /* We used to query real size from OSTs here, but actually
474 this is not needed. For stat() calls size would be updated
475 from subsequent do_revalidate()->ll_inode_revalidate_it() in
477 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
478 Everybody else who needs correct file size would call
479 ll_glimpse_size or some equivalent themselves anyway.
480 Also see bug 7198. */
483 /* Only hash *de if it is unhashed (new dentry).
484 * Atoimc_open may passin hashed dentries for open.
486 if (d_unhashed(*de)) {
487 struct dentry *alias;
489 alias = ll_splice_alias(inode, *de);
491 GOTO(out, rc = PTR_ERR(alias));
494 } else if (!it_disposition(it, DISP_LOOKUP_NEG) &&
495 !it_disposition(it, DISP_OPEN_CREATE)) {
496 /* With DISP_OPEN_CREATE dentry will
497 instantiated in ll_create_it. */
498 LASSERT((*de)->d_inode == NULL);
499 d_instantiate(*de, inode);
502 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
503 /* we have lookup look - unhide dentry */
504 if (bits & MDS_INODELOCK_LOOKUP)
505 d_lustre_revalidate(*de);
506 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
507 /* If file created on server, don't depend on parent UPDATE
508 * lock to unhide it. It is left hidden and next lookup can
509 * find it in ll_splice_alias.
511 /* Check that parent has UPDATE lock. */
512 struct lookup_intent parent_it = {
514 .d.lustre.it_lock_handle = 0 };
515 struct lu_fid fid = ll_i2info(parent)->lli_fid;
517 /* If it is striped directory, get the real stripe parent */
518 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
519 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
520 ll_i2info(parent)->lli_lsm_md,
522 (*de)->d_name.len, &fid);
527 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
529 d_lustre_revalidate(*de);
530 ll_intent_release(&parent_it);
537 if (rc != 0 && it->it_op & IT_OPEN)
538 ll_open_cleanup((*de)->d_sb, request);
543 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
544 struct lookup_intent *it)
546 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
547 struct dentry *save = dentry, *retval;
548 struct ptlrpc_request *req = NULL;
549 struct md_op_data *op_data = NULL;
554 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
555 RETURN(ERR_PTR(-ENAMETOOLONG));
557 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
558 dentry->d_name.len, dentry->d_name.name,
559 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
561 if (d_mountpoint(dentry))
562 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
564 if (it == NULL || it->it_op == IT_GETXATTR)
567 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
568 rc = ll_statahead(parent, &dentry, 0);
570 RETURN(dentry == save ? NULL : dentry);
573 if (it->it_op & IT_CREAT)
574 opc = LUSTRE_OPC_CREATE;
576 opc = LUSTRE_OPC_ANY;
578 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
579 dentry->d_name.len, 0, opc, NULL);
581 RETURN((void *)op_data);
583 /* enforce umask if acl disabled or MDS doesn't support umask */
584 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
585 it->it_create_mode &= ~current_umask();
587 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
588 &ll_md_blocking_ast, 0);
589 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
590 * client does not know which suppgid should be sent to the MDS, or
591 * some other(s) changed the target file's GID after this RPC sent
592 * to the MDS with the suppgid as the original GID, then we should
593 * try again with right suppgid. */
594 if (rc == -EACCES && it->it_op & IT_OPEN &&
595 it_disposition(it, DISP_OPEN_DENY)) {
596 struct mdt_body *body;
598 LASSERT(req != NULL);
600 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
601 if (op_data->op_suppgids[0] == body->mbo_gid ||
602 op_data->op_suppgids[1] == body->mbo_gid ||
603 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
604 GOTO(out, retval = ERR_PTR(-EACCES));
606 fid_zero(&op_data->op_fid2);
607 op_data->op_suppgids[1] = body->mbo_gid;
608 ptlrpc_req_finished(req);
610 ll_intent_release(it);
611 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
612 &ll_md_blocking_ast, 0);
616 GOTO(out, retval = ERR_PTR(rc));
618 rc = ll_lookup_it_finish(req, it, parent, &dentry);
620 ll_intent_release(it);
621 GOTO(out, retval = ERR_PTR(rc));
624 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
625 !S_ISREG(dentry->d_inode->i_mode) &&
626 !S_ISDIR(dentry->d_inode->i_mode)) {
627 ll_release_openhandle(dentry, it);
629 ll_lookup_finish_locks(it, dentry);
631 GOTO(out, retval = (dentry == save) ? NULL : dentry);
634 if (op_data != NULL && !IS_ERR(op_data))
635 ll_finish_md_op_data(op_data);
637 ptlrpc_req_finished(req);
641 #ifdef HAVE_IOP_ATOMIC_OPEN
642 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
645 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
648 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
649 dentry->d_name.len, dentry->d_name.name,
650 PFID(ll_inode2fid(parent)), parent, flags);
652 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
653 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
656 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
660 de = ll_lookup_it(parent, dentry, itp);
663 ll_intent_release(itp);
669 * For cached negative dentry and new dentry, handle lookup/create/open
672 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
673 struct file *file, unsigned open_flags,
674 umode_t mode, int *opened)
676 struct lookup_intent *it;
678 long long lookup_flags = LOOKUP_OPEN;
682 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
683 "open_flags %x, mode %x opened %d\n",
684 dentry->d_name.len, dentry->d_name.name,
685 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
687 OBD_ALLOC(it, sizeof(*it));
692 if (open_flags & O_CREAT) {
693 it->it_op |= IT_CREAT;
694 lookup_flags |= LOOKUP_CREATE;
696 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
697 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
698 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
700 /* Dentry added to dcache tree in ll_lookup_it */
701 de = ll_lookup_it(dir, dentry, it);
708 if (it_disposition(it, DISP_OPEN_CREATE)) {
709 /* Dentry instantiated in ll_create_it. */
710 rc = ll_create_it(dir, dentry, it);
712 /* We dget in ll_splice_alias. */
718 *opened |= FILE_CREATED;
720 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
722 if (S_ISFIFO(dentry->d_inode->i_mode)) {
723 /* We cannot call open here as it would
726 if (it_disposition(it, DISP_ENQ_OPEN_REF))
728 (struct ptlrpc_request *)
729 it->d.lustre.it_data);
730 rc = finish_no_open(file, de);
732 file->private_data = it;
733 rc = finish_open(file, dentry, NULL, opened);
734 /* We dget in ll_splice_alias. finish_open takes
735 * care of dget for fd open.
741 rc = finish_no_open(file, de);
746 ll_intent_release(it);
747 OBD_FREE(it, sizeof(*it));
752 #else /* !HAVE_IOP_ATOMIC_OPEN */
753 static struct lookup_intent *
754 ll_convert_intent(struct open_intent *oit, int lookup_flags)
756 struct lookup_intent *it;
760 return ERR_PTR(-ENOMEM);
762 if (lookup_flags & LOOKUP_OPEN) {
764 if (lookup_flags & LOOKUP_CREATE)
765 it->it_op |= IT_CREAT;
766 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
767 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
768 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
770 it->it_op = IT_GETATTR;
776 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
777 struct nameidata *nd)
782 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
783 struct lookup_intent *it;
785 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
786 it = ll_d2d(dentry)->lld_it;
787 ll_d2d(dentry)->lld_it = NULL;
789 if ((nd->flags & LOOKUP_CREATE) &&
790 !(nd->flags & LOOKUP_OPEN))
793 it = ll_convert_intent(&nd->intent.open, nd->flags);
795 RETURN((struct dentry *)it);
798 de = ll_lookup_it(parent, dentry, it);
801 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
802 if (dentry->d_inode &&
803 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
804 if (S_ISFIFO(dentry->d_inode->i_mode)) {
805 // We cannot call open here as it would
808 (struct ptlrpc_request *)
809 it->d.lustre.it_data);
813 nd->intent.open.file->private_data = it;
814 filp = lookup_instantiate_filp(nd,
820 de = (struct dentry *)filp;
823 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
824 // XXX This can only reliably work on assumption
825 // that there are NO hashed negative dentries.
826 ll_d2d(dentry)->lld_it = it;
827 it = NULL; /* Will be freed in ll_create_nd */
828 /* We absolutely depend on ll_create_nd to be
829 * called to not leak this intent and possible
830 * data attached to it */
835 ll_intent_release(it);
836 OBD_FREE(it, sizeof(*it));
839 de = ll_lookup_it(parent, dentry, NULL);
844 #endif /* HAVE_IOP_ATOMIC_OPEN */
846 /* We depend on "mode" being set with the proper file type/umask by now */
847 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
849 struct inode *inode = NULL;
850 struct ptlrpc_request *request = NULL;
851 struct ll_sb_info *sbi = ll_i2sbi(dir);
855 LASSERT(it && it->d.lustre.it_disposition);
857 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
858 request = it->d.lustre.it_data;
859 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
860 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
862 GOTO(out, inode = ERR_PTR(rc));
864 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
866 /* We asked for a lock on the directory, but were granted a
867 * lock on the inode. Since we finally have an inode pointer,
868 * stuff it in the lock. */
869 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
870 PFID(ll_inode2fid(inode)), inode);
871 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
874 ptlrpc_req_finished(request);
879 * By the time this is called, we already have created the directory cache
880 * entry for the new file, but it is so far negative - it has no inode.
882 * We defer creating the OBD object(s) until open, to keep the intent and
883 * non-intent code paths similar, and also because we do not have the MDS
884 * inode number before calling ll_create_node() (which is needed for LOV),
885 * so we would need to do yet another RPC to the MDS to store the LOV EA
886 * data on the MDS. If needed, we would pass the PACKED lmm as data and
887 * lmm_size in datalen (the MDS still has code which will handle that).
889 * If the create succeeds, we fill in the inode information
890 * with d_instantiate().
892 static int ll_create_it(struct inode *dir, struct dentry *dentry,
893 struct lookup_intent *it)
899 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
900 dentry->d_name.len, dentry->d_name.name,
901 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
903 rc = it_open_error(DISP_OPEN_CREATE, it);
907 inode = ll_create_node(dir, it);
909 RETURN(PTR_ERR(inode));
911 d_instantiate(dentry, inode);
913 rc = ll_init_security(dentry, inode, dir);
920 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
922 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
926 if (body->mbo_valid & OBD_MD_FLMTIME &&
927 body->mbo_mtime > LTIME_S(inode->i_mtime)) {
928 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
929 "\n", PFID(ll_inode2fid(inode)),
930 LTIME_S(inode->i_mtime), body->mbo_mtime);
931 LTIME_S(inode->i_mtime) = body->mbo_mtime;
934 if (body->mbo_valid & OBD_MD_FLCTIME &&
935 body->mbo_ctime > LTIME_S(inode->i_ctime))
936 LTIME_S(inode->i_ctime) = body->mbo_ctime;
939 static int ll_new_node(struct inode *dir, struct dentry *dchild,
940 const char *tgt, umode_t mode, int rdev, __u32 opc)
942 struct qstr *name = &dchild->d_name;
943 struct ptlrpc_request *request = NULL;
944 struct md_op_data *op_data;
945 struct inode *inode = NULL;
946 struct ll_sb_info *sbi = ll_i2sbi(dir);
951 if (unlikely(tgt != NULL))
952 tgt_len = strlen(tgt) + 1;
955 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
956 name->len, 0, opc, NULL);
958 GOTO(err_exit, err = PTR_ERR(op_data));
960 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
961 from_kuid(&init_user_ns, current_fsuid()),
962 from_kgid(&init_user_ns, current_fsgid()),
963 cfs_curproc_cap_pack(), rdev, &request);
964 ll_finish_md_op_data(op_data);
965 if (err < 0 && err != -EREMOTE)
968 /* If the client doesn't know where to create a subdirectory (or
969 * in case of a race that sends the RPC to the wrong MDS), the
970 * MDS will return -EREMOTE and the client will fetch the layout
971 * of the directory, then create the directory on the right MDT. */
972 if (unlikely(err == -EREMOTE)) {
973 struct ll_inode_info *lli = ll_i2info(dir);
974 struct lmv_user_md *lum;
978 ptlrpc_req_finished(request);
981 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
984 /* Update stripe_offset and retry */
985 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
986 } else if (err2 == -ENODATA &&
987 lli->lli_def_stripe_offset != -1) {
988 /* If there are no default stripe EA on the MDT, but the
989 * client has default stripe, then it probably means
990 * default stripe EA has just been deleted. */
991 lli->lli_def_stripe_offset = -1;
996 ptlrpc_req_finished(request);
1001 ll_update_times(request, dir);
1003 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1005 GOTO(err_exit, err);
1007 d_instantiate(dchild, inode);
1009 err = ll_init_security(dchild, inode, dir);
1011 GOTO(err_exit, err);
1015 if (request != NULL)
1016 ptlrpc_req_finished(request);
1021 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1024 struct qstr *name = &dchild->d_name;
1028 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1029 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1032 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1033 mode &= ~current_umask();
1035 switch (mode & S_IFMT) {
1037 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1043 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1054 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1059 #ifdef HAVE_IOP_ATOMIC_OPEN
1061 * Plain create. Intent create is handled in atomic_open.
1063 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1064 umode_t mode, bool want_excl)
1068 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1069 "flags=%u, excl=%d\n", dentry->d_name.len,
1070 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1071 dir, mode, want_excl);
1073 rc = ll_mknod(dir, dentry, mode, 0);
1075 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1077 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1078 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1082 #else /* !HAVE_IOP_ATOMIC_OPEN */
1083 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1084 ll_umode_t mode, struct nameidata *nd)
1086 struct ll_dentry_data *lld = ll_d2d(dentry);
1087 struct lookup_intent *it = NULL;
1094 return ll_mknod(dir, dentry, mode, 0);
1098 /* Was there an error? Propagate it! */
1099 if (it->d.lustre.it_status) {
1100 rc = it->d.lustre.it_status;
1104 rc = ll_create_it(dir, dentry, it);
1105 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1108 nd->intent.open.file->private_data = it;
1109 filp = lookup_instantiate_filp(nd, dentry, NULL);
1115 ll_intent_release(it);
1116 OBD_FREE(it, sizeof(*it));
1119 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1123 #endif /* HAVE_IOP_ATOMIC_OPEN */
1125 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1126 const char *oldpath)
1128 struct qstr *name = &dchild->d_name;
1132 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1133 name->len, name->name, PFID(ll_inode2fid(dir)),
1134 dir, 3000, oldpath);
1136 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1137 LUSTRE_OPC_SYMLINK);
1140 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1145 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1146 struct dentry *new_dentry)
1148 struct inode *src = old_dentry->d_inode;
1149 struct qstr *name = &new_dentry->d_name;
1150 struct ll_sb_info *sbi = ll_i2sbi(dir);
1151 struct ptlrpc_request *request = NULL;
1152 struct md_op_data *op_data;
1156 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1157 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1158 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1160 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1161 0, LUSTRE_OPC_ANY, NULL);
1162 if (IS_ERR(op_data))
1163 RETURN(PTR_ERR(op_data));
1165 err = md_link(sbi->ll_md_exp, op_data, &request);
1166 ll_finish_md_op_data(op_data);
1170 ll_update_times(request, dir);
1171 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1174 ptlrpc_req_finished(request);
1178 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1180 struct qstr *name = &dchild->d_name;
1184 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1185 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1187 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1188 mode &= ~current_umask();
1190 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1192 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1194 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1199 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1201 struct qstr *name = &dchild->d_name;
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 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1210 if (unlikely(d_mountpoint(dchild)))
1213 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1214 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1215 if (IS_ERR(op_data))
1216 RETURN(PTR_ERR(op_data));
1218 if (dchild->d_inode != NULL)
1219 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1221 op_data->op_fid2 = op_data->op_fid3;
1222 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1223 ll_finish_md_op_data(op_data);
1225 ll_update_times(request, dir);
1226 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1229 ptlrpc_req_finished(request);
1236 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1238 struct ptlrpc_request *request = NULL;
1239 struct md_op_data *op_data;
1243 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1244 namelen, name, PFID(ll_inode2fid(dir)), dir);
1246 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1247 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1248 if (IS_ERR(op_data))
1249 RETURN(PTR_ERR(op_data));
1250 op_data->op_cli_flags |= CLI_RM_ENTRY;
1251 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1252 ll_finish_md_op_data(op_data);
1254 ll_update_times(request, dir);
1255 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1258 ptlrpc_req_finished(request);
1262 /* ll_unlink() doesn't update the inode with the new link count.
1263 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if
1264 * there is any lock existing. They will recycle dentries and inodes
1265 * based upon locks too. b=20433 */
1266 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1268 struct qstr *name = &dchild->d_name;
1269 struct ptlrpc_request *request = NULL;
1270 struct md_op_data *op_data;
1273 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1274 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1277 * XXX: unlink bind mountpoint maybe call to here,
1278 * just check it as vfs_unlink does.
1280 if (unlikely(d_mountpoint(dchild)))
1283 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1284 LUSTRE_OPC_ANY, NULL);
1285 if (IS_ERR(op_data))
1286 RETURN(PTR_ERR(op_data));
1288 if (dchild->d_inode != NULL)
1289 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1291 op_data->op_fid2 = op_data->op_fid3;
1292 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1293 ll_finish_md_op_data(op_data);
1297 ll_update_times(request, dir);
1298 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1301 ptlrpc_req_finished(request);
1305 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1306 struct inode *tgt, struct dentry *tgt_dchild)
1308 struct qstr *src_name = &src_dchild->d_name;
1309 struct qstr *tgt_name = &tgt_dchild->d_name;
1310 struct ptlrpc_request *request = NULL;
1311 struct ll_sb_info *sbi = ll_i2sbi(src);
1312 struct md_op_data *op_data;
1315 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1316 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1317 src_name->len, src_name->name,
1318 PFID(ll_inode2fid(src)), src, tgt_name->len,
1319 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1321 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1324 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1325 LUSTRE_OPC_ANY, NULL);
1326 if (IS_ERR(op_data))
1327 RETURN(PTR_ERR(op_data));
1329 if (src_dchild->d_inode != NULL)
1330 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1332 if (tgt_dchild->d_inode != NULL)
1333 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1335 err = md_rename(sbi->ll_md_exp, op_data,
1336 src_name->name, src_name->len,
1337 tgt_name->name, tgt_name->len, &request);
1338 ll_finish_md_op_data(op_data);
1340 ll_update_times(request, src);
1341 ll_update_times(request, tgt);
1342 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1345 ptlrpc_req_finished(request);
1348 d_move(src_dchild, tgt_dchild);
1353 const struct inode_operations ll_dir_inode_operations = {
1355 #ifdef HAVE_IOP_ATOMIC_OPEN
1356 .atomic_open = ll_atomic_open,
1358 .lookup = ll_lookup_nd,
1359 .create = ll_create_nd,
1360 /* We need all these non-raw things for NFSD, to not patch it. */
1361 .unlink = ll_unlink,
1364 .symlink = ll_symlink,
1366 .rename = ll_rename,
1367 .setattr = ll_setattr,
1368 .getattr = ll_getattr,
1369 .permission = ll_inode_permission,
1370 .setxattr = ll_setxattr,
1371 .getxattr = ll_getxattr,
1372 .listxattr = ll_listxattr,
1373 .removexattr = ll_removexattr,
1374 #ifdef HAVE_IOP_GET_ACL
1375 .get_acl = ll_get_acl,
1379 const struct inode_operations ll_special_inode_operations = {
1380 .setattr = ll_setattr,
1381 .getattr = ll_getattr,
1382 .permission = ll_inode_permission,
1383 .setxattr = ll_setxattr,
1384 .getxattr = ll_getxattr,
1385 .listxattr = ll_listxattr,
1386 .removexattr = ll_removexattr,
1387 #ifdef HAVE_IOP_GET_ACL
1388 .get_acl = ll_get_acl,