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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 #include <linux/sched.h>
36 #include <linux/quotaops.h>
37 #include <linux/highmem.h>
38 #include <linux/pagemap.h>
39 #include <linux/security.h>
40 #include <linux/user_namespace.h>
41 #ifdef HAVE_UIDGID_HEADER
42 # include <linux/uidgid.h>
45 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <lustre_dlm.h>
50 #include "llite_internal.h"
52 static int ll_create_it(struct inode *dir, struct dentry *dentry,
53 struct lookup_intent *it,
54 void *secctx, __u32 secctxlen);
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 rc = cl_file_inode_init(inode, md);
127 /* Let's clear directory lsm here, otherwise
128 * make_bad_inode() will reset the inode mode
129 * to regular, then ll_clear_inode will not
130 * be able to clear lsm_md */
131 if (S_ISDIR(inode->i_mode))
132 ll_dir_clear_lsm_md(inode);
133 make_bad_inode(inode);
134 unlock_new_inode(inode);
138 inode_has_no_xattr(inode);
139 unlock_new_inode(inode);
141 } else if (is_bad_inode(inode)) {
143 inode = ERR_PTR(-ESTALE);
144 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
145 rc = ll_update_inode(inode, md);
146 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
147 PFID(&md->body->mbo_fid1), inode, rc);
149 if (S_ISDIR(inode->i_mode))
150 ll_dir_clear_lsm_md(inode);
159 static void ll_invalidate_negative_children(struct inode *dir)
161 struct dentry *dentry, *tmp_subdir;
162 DECLARE_LL_D_HLIST_NODE_PTR(p);
165 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
166 spin_lock(&dentry->d_lock);
167 if (!list_empty(&dentry->d_subdirs)) {
168 struct dentry *child;
170 list_for_each_entry_safe(child, tmp_subdir,
173 if (child->d_inode == NULL)
174 d_lustre_invalidate(child, 1);
177 spin_unlock(&dentry->d_lock);
179 ll_unlock_dcache(dir);
182 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
184 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
187 int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
190 struct ll_inode_info *lli = ll_i2info(inode);
191 struct cl_layout clt = { .cl_layout_gen = 0, };
201 env = cl_env_get(&refcheck);
203 RETURN(PTR_ERR(env));
205 rc = cl_object_layout_get(env, lli->lli_clob, &clt);
207 CDEBUG(D_INODE, "Cannot get layout for "DFID"\n",
208 PFID(ll_inode2fid(inode)));
210 } else if (clt.cl_size == 0 || clt.cl_dom_comp_size == 0) {
211 CDEBUG(D_INODE, "DOM lock without DOM layout for "DFID"\n",
212 PFID(ll_inode2fid(inode)));
214 enum cl_fsync_mode mode;
215 loff_t end = clt.cl_dom_comp_size - 1;
217 mode = ldlm_is_discard_data(lock) ?
218 CL_FSYNC_DISCARD : CL_FSYNC_LOCAL;
219 rc = cl_sync_file_range(inode, 0, end, mode, 1);
220 truncate_inode_pages_range(inode->i_mapping, 0, end);
222 cl_env_put(env, &refcheck);
226 void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
228 struct inode *inode = ll_inode_from_resource_lock(lock);
229 __u64 bits = to_cancel;
235 if (!fid_res_name_eq(ll_inode2fid(inode),
236 &lock->l_resource->lr_name)) {
237 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
238 PFID(ll_inode2fid(inode)), inode);
242 if (bits & MDS_INODELOCK_XATTR) {
243 if (S_ISDIR(inode->i_mode))
244 ll_i2info(inode)->lli_def_stripe_offset = -1;
245 ll_xattr_cache_destroy(inode);
246 bits &= ~MDS_INODELOCK_XATTR;
249 /* For OPEN locks we differentiate between lock modes
250 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
251 if (bits & MDS_INODELOCK_OPEN)
252 ll_have_md_lock(inode, &bits, lock->l_req_mode);
254 if (bits & MDS_INODELOCK_OPEN) {
257 switch (lock->l_req_mode) {
268 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
272 ll_md_real_close(inode, fmode);
274 bits &= ~MDS_INODELOCK_OPEN;
277 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
278 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
280 ll_have_md_lock(inode, &bits, LCK_MINMODE);
282 if (bits & MDS_INODELOCK_DOM) {
283 rc = ll_dom_lock_cancel(inode, lock);
285 CDEBUG(D_INODE, "cannot flush DoM data "
287 PFID(ll_inode2fid(inode)), rc);
288 lock_res_and_lock(lock);
289 ldlm_set_kms_ignore(lock);
290 unlock_res_and_lock(lock);
293 if (bits & MDS_INODELOCK_LAYOUT) {
294 struct cl_object_conf conf = {
295 .coc_opc = OBJECT_CONF_INVALIDATE,
299 rc = ll_layout_conf(inode, &conf);
301 CDEBUG(D_INODE, "cannot invalidate layout of "
303 PFID(ll_inode2fid(inode)), rc);
306 if (bits & MDS_INODELOCK_UPDATE) {
307 struct ll_inode_info *lli = ll_i2info(inode);
309 lli->lli_update_atime = 1;
312 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
313 struct ll_inode_info *lli = ll_i2info(inode);
315 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
316 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
317 lli, PFID(&lli->lli_pfid));
318 truncate_inode_pages(inode->i_mapping, 0);
320 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
321 struct inode *master_inode = NULL;
324 /* This is slave inode, since all of the child dentry
325 * is connected on the master inode, so we have to
326 * invalidate the negative children on master inode */
327 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
328 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
330 hash = cl_fid_build_ino(&lli->lli_pfid,
331 ll_need_32bit_api(ll_i2sbi(inode)));
333 /* Do not lookup the inode with ilookup5, otherwise
334 * it will cause dead lock,
335 * 1. Client1 send chmod req to the MDT0, then on MDT0,
336 * it enqueues master and all of its slaves lock,
337 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
338 * master and stripe0 lock, it will send the enqueue
339 * req (for stripe1) to MDT1, then MDT1 finds the lock
340 * has been granted to client2. Then MDT1 sends blocking
342 * 2. At the same time, client2 tries to unlink
343 * the striped dir (rm -rf striped_dir), and during
344 * lookup, it will hold the master inode of the striped
345 * directory, whose inode state is NEW, then tries to
346 * revalidate all of its slaves, (ll_prep_inode()->
347 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
348 * And it will be blocked on the server side because
350 * 3. Then the client get the blocking_ast req, cancel
351 * the lock, but being blocked if using ->ilookup5()),
352 * because master inode state is NEW. */
353 master_inode = ilookup5_nowait(inode->i_sb, hash,
354 ll_test_inode_by_fid,
355 (void *)&lli->lli_pfid);
357 ll_invalidate_negative_children(master_inode);
361 ll_invalidate_negative_children(inode);
365 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
366 inode->i_sb->s_root != NULL &&
367 inode != inode->i_sb->s_root->d_inode)
368 ll_invalidate_aliases(inode);
373 /* Check if the given lock may be downgraded instead of canceling and
374 * that convert is really needed. */
375 int ll_md_need_convert(struct ldlm_lock *lock)
378 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
379 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
380 enum ldlm_mode mode = LCK_MINMODE;
382 if (!wanted || !bits || ldlm_is_cancel(lock))
385 /* do not convert locks other than DOM for now */
386 if (!((bits | wanted) & MDS_INODELOCK_DOM))
389 /* We may have already remaining bits in some other lock so
390 * lock convert will leave us just extra lock for the same bit.
391 * Check if client has other lock with the same bits and the same
392 * or lower mode and don't convert if any.
394 switch (lock->l_req_mode) {
406 /* do not convert other modes */
410 /* is lock is too old to be converted? */
411 lock_res_and_lock(lock);
412 if (ktime_after(ktime_get(),
413 ktime_add(lock->l_last_used,
414 ktime_set(10, 0)))) {
415 unlock_res_and_lock(lock);
418 unlock_res_and_lock(lock);
420 inode = ll_inode_from_resource_lock(lock);
421 ll_have_md_lock(inode, &bits, mode);
426 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
427 void *data, int flag)
429 struct lustre_handle lockh;
430 __u64 bits = lock->l_policy_data.l_inodebits.bits;
436 case LDLM_CB_BLOCKING:
438 __u64 cancel_flags = LCF_ASYNC;
440 if (ll_md_need_convert(lock)) {
441 cancel_flags |= LCF_CONVERT;
442 /* For lock convert some cancel actions may require
443 * this lock with non-dropped canceled bits, e.g. page
444 * flush for DOM lock. So call ll_lock_cancel_bits()
445 * here while canceled bits are still set.
447 bits = lock->l_policy_data.l_inodebits.cancel_bits;
448 if (bits & MDS_INODELOCK_DOM)
449 ll_lock_cancel_bits(lock, MDS_INODELOCK_DOM);
451 ldlm_lock2handle(lock, &lockh);
452 rc = ldlm_cli_cancel(&lockh, cancel_flags);
454 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
459 case LDLM_CB_CANCELING:
460 if (ldlm_is_converting(lock)) {
461 /* this is called on already converted lock, so
462 * ibits has remained bits only and cancel_bits
463 * are bits that were dropped.
464 * Note that DOM lock is handled prior lock convert
465 * and is excluded here.
467 bits = lock->l_policy_data.l_inodebits.cancel_bits &
470 LASSERT(ldlm_is_canceling(lock));
472 ll_lock_cancel_bits(lock, bits);
481 __u32 ll_i2suppgid(struct inode *i)
483 if (in_group_p(i->i_gid))
484 return (__u32)from_kgid(&init_user_ns, i->i_gid);
486 return (__u32) __kgid_val(INVALID_GID);
489 /* Pack the required supplementary groups into the supplied groups array.
490 * If we don't need to use the groups from the target inode(s) then we
491 * instead pack one or more groups from the user's supplementary group
492 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
493 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
496 LASSERT(suppgids != NULL);
498 suppgids[0] = ll_i2suppgid(i1);
501 suppgids[1] = ll_i2suppgid(i2);
507 * try to reuse three types of dentry:
508 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
509 * by concurrent .revalidate).
510 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
511 * be cleared by others calling d_lustre_revalidate).
512 * 3. DISCONNECTED alias.
514 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
516 struct dentry *alias, *discon_alias, *invalid_alias;
517 DECLARE_LL_D_HLIST_NODE_PTR(p);
519 if (ll_d_hlist_empty(&inode->i_dentry))
522 discon_alias = invalid_alias = NULL;
524 ll_lock_dcache(inode);
525 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
526 LASSERT(alias != dentry);
528 spin_lock(&alias->d_lock);
529 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
530 S_ISDIR(inode->i_mode))
531 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
532 discon_alias = alias;
533 else if (alias->d_parent == dentry->d_parent &&
534 alias->d_name.hash == dentry->d_name.hash &&
535 alias->d_name.len == dentry->d_name.len &&
536 memcmp(alias->d_name.name, dentry->d_name.name,
537 dentry->d_name.len) == 0)
538 invalid_alias = alias;
539 spin_unlock(&alias->d_lock);
544 alias = invalid_alias ?: discon_alias ?: NULL;
546 spin_lock(&alias->d_lock);
548 spin_unlock(&alias->d_lock);
550 ll_unlock_dcache(inode);
556 * Similar to d_splice_alias(), but lustre treats invalid alias
557 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
559 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
565 new = ll_find_alias(inode, de);
575 "Reuse dentry %p inode %p refc %d flags %#x\n",
576 new, new->d_inode, ll_d_count(new), new->d_flags);
584 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
585 de, de->d_inode, ll_d_count(de), de->d_flags);
589 static int ll_lookup_it_finish(struct ptlrpc_request *request,
590 struct lookup_intent *it,
591 struct inode *parent, struct dentry **de)
593 struct inode *inode = NULL;
596 struct dentry *alias;
599 /* NB 1 request reference will be taken away by ll_intent_lock()
601 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
603 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
604 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
608 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
610 /* We used to query real size from OSTs here, but actually
611 this is not needed. For stat() calls size would be updated
612 from subsequent do_revalidate()->ll_inode_revalidate_it() in
614 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
615 Everybody else who needs correct file size would call
616 ll_glimpse_size or some equivalent themselves anyway.
617 Also see bug 7198. */
620 /* Only hash *de if it is unhashed (new dentry).
621 * Atoimc_open may passin hashed dentries for open.
623 alias = ll_splice_alias(inode, *de);
625 GOTO(out, rc = PTR_ERR(alias));
629 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
630 /* we have lookup look - unhide dentry */
631 if (bits & MDS_INODELOCK_LOOKUP)
632 d_lustre_revalidate(*de);
633 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
634 /* If file created on server, don't depend on parent UPDATE
635 * lock to unhide it. It is left hidden and next lookup can
636 * find it in ll_splice_alias.
638 /* Check that parent has UPDATE lock. */
639 struct lookup_intent parent_it = {
641 .it_lock_handle = 0 };
642 struct lu_fid fid = ll_i2info(parent)->lli_fid;
644 /* If it is striped directory, get the real stripe parent */
645 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
646 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
647 ll_i2info(parent)->lli_lsm_md,
649 (*de)->d_name.len, &fid);
654 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
656 d_lustre_revalidate(*de);
657 ll_intent_release(&parent_it);
664 if (rc != 0 && it->it_op & IT_OPEN)
665 ll_open_cleanup((*de)->d_sb, request);
670 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
671 struct lookup_intent *it,
672 void **secctx, __u32 *secctxlen)
674 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
675 struct dentry *save = dentry, *retval;
676 struct ptlrpc_request *req = NULL;
677 struct md_op_data *op_data = NULL;
682 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
683 RETURN(ERR_PTR(-ENAMETOOLONG));
685 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
686 dentry->d_name.len, dentry->d_name.name,
687 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
689 if (d_mountpoint(dentry))
690 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
692 if (it == NULL || it->it_op == IT_GETXATTR)
695 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
696 rc = ll_statahead(parent, &dentry, 0);
698 RETURN(dentry == save ? NULL : dentry);
701 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
702 dentry->d_sb->s_flags & MS_RDONLY)
703 RETURN(ERR_PTR(-EROFS));
705 if (it->it_op & IT_CREAT)
706 opc = LUSTRE_OPC_CREATE;
708 opc = LUSTRE_OPC_ANY;
710 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
711 dentry->d_name.len, 0, opc, NULL);
713 GOTO(out, retval = ERR_CAST(op_data));
715 /* enforce umask if acl disabled or MDS doesn't support umask */
716 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
717 it->it_create_mode &= ~current_umask();
719 if (it->it_op & IT_CREAT &&
720 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
721 rc = ll_dentry_init_security(dentry, it->it_create_mode,
723 &op_data->op_file_secctx_name,
724 &op_data->op_file_secctx,
725 &op_data->op_file_secctx_size);
727 GOTO(out, retval = ERR_PTR(rc));
729 *secctx = op_data->op_file_secctx;
730 if (secctxlen != NULL)
731 *secctxlen = op_data->op_file_secctx_size;
734 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
735 &ll_md_blocking_ast, 0);
736 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
737 * client does not know which suppgid should be sent to the MDS, or
738 * some other(s) changed the target file's GID after this RPC sent
739 * to the MDS with the suppgid as the original GID, then we should
740 * try again with right suppgid. */
741 if (rc == -EACCES && it->it_op & IT_OPEN &&
742 it_disposition(it, DISP_OPEN_DENY)) {
743 struct mdt_body *body;
745 LASSERT(req != NULL);
747 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
748 if (op_data->op_suppgids[0] == body->mbo_gid ||
749 op_data->op_suppgids[1] == body->mbo_gid ||
750 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
751 GOTO(out, retval = ERR_PTR(-EACCES));
753 fid_zero(&op_data->op_fid2);
754 op_data->op_suppgids[1] = body->mbo_gid;
755 ptlrpc_req_finished(req);
757 ll_intent_release(it);
758 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
759 &ll_md_blocking_ast, 0);
763 GOTO(out, retval = ERR_PTR(rc));
765 rc = ll_lookup_it_finish(req, it, parent, &dentry);
767 ll_intent_release(it);
768 GOTO(out, retval = ERR_PTR(rc));
771 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
772 !S_ISREG(dentry->d_inode->i_mode) &&
773 !S_ISDIR(dentry->d_inode->i_mode)) {
774 ll_release_openhandle(dentry, it);
776 ll_lookup_finish_locks(it, dentry);
778 GOTO(out, retval = (dentry == save) ? NULL : dentry);
781 if (op_data != NULL && !IS_ERR(op_data)) {
782 if (secctx != NULL && secctxlen != NULL) {
783 /* caller needs sec ctx info, so reset it in op_data to
784 * prevent it from being freed */
785 op_data->op_file_secctx = NULL;
786 op_data->op_file_secctx_size = 0;
788 ll_finish_md_op_data(op_data);
791 ptlrpc_req_finished(req);
795 #ifdef HAVE_IOP_ATOMIC_OPEN
796 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
799 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
802 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
803 dentry->d_name.len, dentry->d_name.name,
804 PFID(ll_inode2fid(parent)), parent, flags);
807 * Optimize away (CREATE && !OPEN). Let .create handle the race.
808 * but only if we have write permissions there, otherwise we need
809 * to proceed with lookup. LU-4185
811 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
812 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
815 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
819 de = ll_lookup_it(parent, dentry, itp, NULL, NULL);
822 ll_intent_release(itp);
828 * For cached negative dentry and new dentry, handle lookup/create/open
831 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
832 struct file *file, unsigned open_flags,
833 umode_t mode, int *opened)
835 struct lookup_intent *it;
837 long long lookup_flags = LOOKUP_OPEN;
843 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
844 "open_flags %x, mode %x opened %d\n",
845 dentry->d_name.len, dentry->d_name.name,
846 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
848 /* Only negative dentries enter here */
849 LASSERT(dentry->d_inode == NULL);
851 if (!d_unhashed(dentry)) {
852 /* A valid negative dentry that just passed revalidation,
853 * there's little point to try and open it server-side,
854 * even though there's a minuscule chance it might succeed.
855 * Either way it's a valid race to just return -ENOENT here.
857 if (!(open_flags & O_CREAT))
860 /* Otherwise we just unhash it to be rehashed afresh via
861 * lookup if necessary
866 OBD_ALLOC(it, sizeof(*it));
871 if (open_flags & O_CREAT) {
872 it->it_op |= IT_CREAT;
873 lookup_flags |= LOOKUP_CREATE;
875 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
876 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
877 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
879 /* Dentry added to dcache tree in ll_lookup_it */
880 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen);
886 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
889 if (it_disposition(it, DISP_OPEN_CREATE)) {
890 /* Dentry instantiated in ll_create_it. */
891 rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
892 security_release_secctx(secctx, secctxlen);
894 /* We dget in ll_splice_alias. */
900 *opened |= FILE_CREATED;
902 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
904 if (S_ISFIFO(dentry->d_inode->i_mode)) {
905 /* We cannot call open here as it might
906 * deadlock. This case is unreachable in
907 * practice because of OBD_CONNECT_NODEVOH. */
908 rc = finish_no_open(file, de);
910 file->private_data = it;
911 rc = finish_open(file, dentry, NULL, opened);
912 /* We dget in ll_splice_alias. finish_open takes
913 * care of dget for fd open.
919 rc = finish_no_open(file, de);
924 ll_intent_release(it);
925 OBD_FREE(it, sizeof(*it));
930 #else /* !HAVE_IOP_ATOMIC_OPEN */
931 static struct lookup_intent *
932 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
934 struct lookup_intent *it;
938 return ERR_PTR(-ENOMEM);
940 if (lookup_flags & LOOKUP_OPEN) {
942 /* Avoid file creation for ro bind mount point(is_readonly) */
943 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
944 it->it_op |= IT_CREAT;
945 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
946 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
947 ~(is_readonly ? O_CREAT : 0));
948 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
950 it->it_op = IT_GETATTR;
956 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
957 struct nameidata *nd)
962 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
963 struct lookup_intent *it;
965 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
966 it = ll_d2d(dentry)->lld_it;
967 ll_d2d(dentry)->lld_it = NULL;
970 * Optimize away (CREATE && !OPEN). Let .create handle
971 * the race. But only if we have write permissions
972 * there, otherwise we need to proceed with lookup.
975 if ((nd->flags & LOOKUP_CREATE) &&
976 !(nd->flags & LOOKUP_OPEN) &&
977 (inode_permission(parent,
978 MAY_WRITE | MAY_EXEC) == 0))
981 it = ll_convert_intent(&nd->intent.open, nd->flags,
982 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
983 (nd->path.mnt->mnt_sb->s_flags & MS_RDONLY));
985 RETURN((struct dentry *)it);
988 de = ll_lookup_it(parent, dentry, it, NULL, NULL);
991 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
992 if (dentry->d_inode &&
993 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
994 if (S_ISFIFO(dentry->d_inode->i_mode)) {
995 /* We cannot call open here as it might
996 * deadlock. This case is unreachable in
997 * practice because of
998 * OBD_CONNECT_NODEVOH. */
1002 nd->intent.open.file->private_data = it;
1003 filp = lookup_instantiate_filp(nd,
1009 de = (struct dentry *)filp;
1012 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1013 /* XXX This can only reliably work on assumption
1014 * that there are NO hashed negative dentries.*/
1015 ll_d2d(dentry)->lld_it = it;
1016 it = NULL; /* Will be freed in ll_create_nd */
1017 /* We absolutely depend on ll_create_nd to be
1018 * called to not leak this intent and possible
1019 * data attached to it */
1024 ll_intent_release(it);
1025 OBD_FREE(it, sizeof(*it));
1028 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL);
1033 #endif /* HAVE_IOP_ATOMIC_OPEN */
1035 /* We depend on "mode" being set with the proper file type/umask by now */
1036 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1038 struct inode *inode = NULL;
1039 struct ptlrpc_request *request = NULL;
1040 struct ll_sb_info *sbi = ll_i2sbi(dir);
1044 LASSERT(it && it->it_disposition);
1046 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1047 request = it->it_request;
1048 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1049 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1051 GOTO(out, inode = ERR_PTR(rc));
1053 /* Pause to allow for a race with concurrent access by fid */
1054 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1056 /* We asked for a lock on the directory, but were granted a
1057 * lock on the inode. Since we finally have an inode pointer,
1058 * stuff it in the lock. */
1059 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1060 PFID(ll_inode2fid(inode)), inode);
1061 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1064 ptlrpc_req_finished(request);
1069 * By the time this is called, we already have created the directory cache
1070 * entry for the new file, but it is so far negative - it has no inode.
1072 * We defer creating the OBD object(s) until open, to keep the intent and
1073 * non-intent code paths similar, and also because we do not have the MDS
1074 * inode number before calling ll_create_node() (which is needed for LOV),
1075 * so we would need to do yet another RPC to the MDS to store the LOV EA
1076 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1077 * lmm_size in datalen (the MDS still has code which will handle that).
1079 * If the create succeeds, we fill in the inode information
1080 * with d_instantiate().
1082 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1083 struct lookup_intent *it,
1084 void *secctx, __u32 secctxlen)
1086 struct inode *inode;
1090 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1091 dentry->d_name.len, dentry->d_name.name,
1092 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1094 rc = it_open_error(DISP_OPEN_CREATE, it);
1098 inode = ll_create_node(dir, it);
1100 RETURN(PTR_ERR(inode));
1102 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1105 /* must be done before d_instantiate, because it calls
1106 * security_d_instantiate, which means a getxattr if security
1107 * context is not set yet */
1108 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1109 inode_unlock(inode);
1114 d_instantiate(dentry, inode);
1116 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1117 rc = ll_inode_init_security(dentry, inode, dir);
1125 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1127 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1131 if (body->mbo_valid & OBD_MD_FLMTIME &&
1132 body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1133 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to %llu"
1134 "\n", PFID(ll_inode2fid(inode)),
1135 LTIME_S(inode->i_mtime), body->mbo_mtime);
1136 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1139 if (body->mbo_valid & OBD_MD_FLCTIME &&
1140 body->mbo_ctime > LTIME_S(inode->i_ctime))
1141 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1144 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1145 const char *tgt, umode_t mode, int rdev, __u32 opc)
1147 struct qstr *name = &dchild->d_name;
1148 struct ptlrpc_request *request = NULL;
1149 struct md_op_data *op_data;
1150 struct inode *inode = NULL;
1151 struct ll_sb_info *sbi = ll_i2sbi(dir);
1156 if (unlikely(tgt != NULL))
1157 tgt_len = strlen(tgt) + 1;
1160 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1161 name->len, 0, opc, NULL);
1162 if (IS_ERR(op_data))
1163 GOTO(err_exit, err = PTR_ERR(op_data));
1165 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1166 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1167 &op_data->op_file_secctx_name,
1168 &op_data->op_file_secctx,
1169 &op_data->op_file_secctx_size);
1171 GOTO(err_exit, err);
1174 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1175 from_kuid(&init_user_ns, current_fsuid()),
1176 from_kgid(&init_user_ns, current_fsgid()),
1177 cfs_curproc_cap_pack(), rdev, &request);
1178 if (err < 0 && err != -EREMOTE)
1179 GOTO(err_exit, err);
1181 /* If the client doesn't know where to create a subdirectory (or
1182 * in case of a race that sends the RPC to the wrong MDS), the
1183 * MDS will return -EREMOTE and the client will fetch the layout
1184 * of the directory, then create the directory on the right MDT. */
1185 if (unlikely(err == -EREMOTE)) {
1186 struct ll_inode_info *lli = ll_i2info(dir);
1187 struct lmv_user_md *lum;
1191 ptlrpc_req_finished(request);
1194 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1195 OBD_MD_DEFAULT_MEA);
1197 /* Update stripe_offset and retry */
1198 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1199 } else if (err2 == -ENODATA &&
1200 lli->lli_def_stripe_offset != -1) {
1201 /* If there are no default stripe EA on the MDT, but the
1202 * client has default stripe, then it probably means
1203 * default stripe EA has just been deleted. */
1204 lli->lli_def_stripe_offset = -1;
1206 GOTO(err_exit, err);
1209 ptlrpc_req_finished(request);
1211 ll_finish_md_op_data(op_data);
1215 ll_update_times(request, dir);
1217 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1219 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1221 GOTO(err_exit, err);
1223 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1225 /* must be done before d_instantiate, because it calls
1226 * security_d_instantiate, which means a getxattr if security
1227 * context is not set yet */
1228 err = security_inode_notifysecctx(inode,
1229 op_data->op_file_secctx,
1230 op_data->op_file_secctx_size);
1231 inode_unlock(inode);
1233 GOTO(err_exit, err);
1236 d_instantiate(dchild, inode);
1238 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1239 err = ll_inode_init_security(dchild, inode, dir);
1241 GOTO(err_exit, err);
1246 if (request != NULL)
1247 ptlrpc_req_finished(request);
1249 if (!IS_ERR_OR_NULL(op_data))
1250 ll_finish_md_op_data(op_data);
1255 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1258 struct qstr *name = &dchild->d_name;
1262 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1263 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1266 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1267 mode &= ~current_umask();
1269 switch (mode & S_IFMT) {
1271 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1277 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1288 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1293 #ifdef HAVE_IOP_ATOMIC_OPEN
1295 * Plain create. Intent create is handled in atomic_open.
1297 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1298 umode_t mode, bool want_excl)
1302 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1304 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1305 "flags=%u, excl=%d\n", dentry->d_name.len,
1306 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1307 dir, mode, want_excl);
1309 /* Using mknod(2) to create a regular file is designed to not recognize
1310 * volatile file name, so we use ll_mknod() here. */
1311 rc = ll_mknod(dir, dentry, mode, 0);
1313 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1315 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1316 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1320 #else /* !HAVE_IOP_ATOMIC_OPEN */
1321 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1322 ll_umode_t mode, struct nameidata *nd)
1324 struct ll_dentry_data *lld = ll_d2d(dentry);
1325 struct lookup_intent *it = NULL;
1328 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1334 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1335 * so that volatile file name is recoginized.
1336 * Mknod(2), however, is designed to not recognize volatile
1337 * file name to avoid inode leak under orphan directory until
1339 return ll_new_node(dir, dentry, NULL, mode, 0,
1345 /* Was there an error? Propagate it! */
1346 if (it->it_status) {
1351 rc = ll_create_it(dir, dentry, it, NULL, 0);
1352 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1355 nd->intent.open.file->private_data = it;
1356 filp = lookup_instantiate_filp(nd, dentry, NULL);
1362 ll_intent_release(it);
1363 OBD_FREE(it, sizeof(*it));
1366 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1370 #endif /* HAVE_IOP_ATOMIC_OPEN */
1372 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1373 const char *oldpath)
1375 struct qstr *name = &dchild->d_name;
1379 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1380 name->len, name->name, PFID(ll_inode2fid(dir)),
1381 dir, 3000, oldpath);
1383 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1384 LUSTRE_OPC_SYMLINK);
1387 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1392 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1393 struct dentry *new_dentry)
1395 struct inode *src = old_dentry->d_inode;
1396 struct qstr *name = &new_dentry->d_name;
1397 struct ll_sb_info *sbi = ll_i2sbi(dir);
1398 struct ptlrpc_request *request = NULL;
1399 struct md_op_data *op_data;
1403 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1404 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1405 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1407 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1408 0, LUSTRE_OPC_ANY, NULL);
1409 if (IS_ERR(op_data))
1410 RETURN(PTR_ERR(op_data));
1412 err = md_link(sbi->ll_md_exp, op_data, &request);
1413 ll_finish_md_op_data(op_data);
1417 ll_update_times(request, dir);
1418 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1421 ptlrpc_req_finished(request);
1425 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1427 struct qstr *name = &dchild->d_name;
1431 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1432 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1434 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1435 mode &= ~current_umask();
1437 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1439 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1441 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1446 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1448 struct qstr *name = &dchild->d_name;
1449 struct ptlrpc_request *request = NULL;
1450 struct md_op_data *op_data;
1454 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1455 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1457 if (unlikely(d_mountpoint(dchild)))
1460 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1461 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1462 if (IS_ERR(op_data))
1463 RETURN(PTR_ERR(op_data));
1465 if (dchild->d_inode != NULL)
1466 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1468 op_data->op_fid2 = op_data->op_fid3;
1469 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1470 ll_finish_md_op_data(op_data);
1472 ll_update_times(request, dir);
1473 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1476 ptlrpc_req_finished(request);
1483 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1485 struct ptlrpc_request *request = NULL;
1486 struct md_op_data *op_data;
1490 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1491 namelen, name, PFID(ll_inode2fid(dir)), dir);
1493 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1494 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1495 if (IS_ERR(op_data))
1496 RETURN(PTR_ERR(op_data));
1497 op_data->op_cli_flags |= CLI_RM_ENTRY;
1498 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1499 ll_finish_md_op_data(op_data);
1501 ll_update_times(request, dir);
1502 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1505 ptlrpc_req_finished(request);
1509 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1511 struct qstr *name = &dchild->d_name;
1512 struct ptlrpc_request *request = NULL;
1513 struct md_op_data *op_data;
1514 struct mdt_body *body;
1517 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1518 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1521 * XXX: unlink bind mountpoint maybe call to here,
1522 * just check it as vfs_unlink does.
1524 if (unlikely(d_mountpoint(dchild)))
1527 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1528 LUSTRE_OPC_ANY, NULL);
1529 if (IS_ERR(op_data))
1530 RETURN(PTR_ERR(op_data));
1532 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1534 op_data->op_fid2 = op_data->op_fid3;
1535 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1536 ll_finish_md_op_data(op_data);
1541 * The server puts attributes in on the last unlink, use them to update
1542 * the link count so the inode can be freed immediately.
1544 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1545 if (body->mbo_valid & OBD_MD_FLNLINK)
1546 set_nlink(dchild->d_inode, body->mbo_nlink);
1548 ll_update_times(request, dir);
1549 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1552 ptlrpc_req_finished(request);
1556 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1557 struct inode *tgt, struct dentry *tgt_dchild
1558 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1559 , unsigned int flags
1563 struct qstr *src_name = &src_dchild->d_name;
1564 struct qstr *tgt_name = &tgt_dchild->d_name;
1565 struct ptlrpc_request *request = NULL;
1566 struct ll_sb_info *sbi = ll_i2sbi(src);
1567 struct md_op_data *op_data;
1571 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1576 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1577 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1578 src_name->len, src_name->name,
1579 PFID(ll_inode2fid(src)), src, tgt_name->len,
1580 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1582 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1585 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1586 LUSTRE_OPC_ANY, NULL);
1587 if (IS_ERR(op_data))
1588 RETURN(PTR_ERR(op_data));
1590 if (src_dchild->d_inode != NULL)
1591 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1593 if (tgt_dchild->d_inode != NULL)
1594 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1596 err = md_rename(sbi->ll_md_exp, op_data,
1597 src_name->name, src_name->len,
1598 tgt_name->name, tgt_name->len, &request);
1599 ll_finish_md_op_data(op_data);
1601 ll_update_times(request, src);
1602 ll_update_times(request, tgt);
1603 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1606 ptlrpc_req_finished(request);
1609 d_move(src_dchild, tgt_dchild);
1614 const struct inode_operations ll_dir_inode_operations = {
1616 #ifdef HAVE_IOP_ATOMIC_OPEN
1617 .atomic_open = ll_atomic_open,
1619 .lookup = ll_lookup_nd,
1620 .create = ll_create_nd,
1621 /* We need all these non-raw things for NFSD, to not patch it. */
1622 .unlink = ll_unlink,
1625 .symlink = ll_symlink,
1627 .rename = ll_rename,
1628 .setattr = ll_setattr,
1629 .getattr = ll_getattr,
1630 .permission = ll_inode_permission,
1631 #ifdef HAVE_IOP_XATTR
1632 .setxattr = ll_setxattr,
1633 .getxattr = ll_getxattr,
1634 .removexattr = ll_removexattr,
1636 .listxattr = ll_listxattr,
1637 #ifdef HAVE_IOP_GET_ACL
1638 .get_acl = ll_get_acl,
1640 #ifdef HAVE_IOP_SET_ACL
1641 .set_acl = ll_set_acl,
1645 const struct inode_operations ll_special_inode_operations = {
1646 .setattr = ll_setattr,
1647 .getattr = ll_getattr,
1648 .permission = ll_inode_permission,
1649 #ifdef HAVE_IOP_XATTR
1650 .setxattr = ll_setxattr,
1651 .getxattr = ll_getxattr,
1652 .removexattr = ll_removexattr,
1654 .listxattr = ll_listxattr,
1655 #ifdef HAVE_IOP_GET_ACL
1656 .get_acl = ll_get_acl,
1658 #ifdef HAVE_IOP_SET_ACL
1659 .set_acl = ll_set_acl,