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, };
198 if (!lli->lli_clob) {
199 /* due to DoM read on open, there may exist pages for Lustre
200 * regular file even though cl_object is not set up yet. */
201 truncate_inode_pages(inode->i_mapping, 0);
205 env = cl_env_get(&refcheck);
207 RETURN(PTR_ERR(env));
209 rc = cl_object_layout_get(env, lli->lli_clob, &clt);
211 CDEBUG(D_INODE, "Cannot get layout for "DFID"\n",
212 PFID(ll_inode2fid(inode)));
214 } else if (clt.cl_size == 0 || clt.cl_dom_comp_size == 0) {
215 CDEBUG(D_INODE, "DOM lock without DOM layout for "DFID"\n",
216 PFID(ll_inode2fid(inode)));
218 enum cl_fsync_mode mode;
219 loff_t end = clt.cl_dom_comp_size - 1;
221 mode = ldlm_is_discard_data(lock) ?
222 CL_FSYNC_DISCARD : CL_FSYNC_LOCAL;
223 rc = cl_sync_file_range(inode, 0, end, mode, 1);
224 truncate_inode_pages_range(inode->i_mapping, 0, end);
226 cl_env_put(env, &refcheck);
230 void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
232 struct inode *inode = ll_inode_from_resource_lock(lock);
233 __u64 bits = to_cancel;
239 if (!fid_res_name_eq(ll_inode2fid(inode),
240 &lock->l_resource->lr_name)) {
241 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
242 PFID(ll_inode2fid(inode)), inode);
246 if (bits & MDS_INODELOCK_XATTR) {
247 if (S_ISDIR(inode->i_mode))
248 ll_i2info(inode)->lli_def_stripe_offset = -1;
249 ll_xattr_cache_destroy(inode);
250 bits &= ~MDS_INODELOCK_XATTR;
253 /* For OPEN locks we differentiate between lock modes
254 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
255 if (bits & MDS_INODELOCK_OPEN)
256 ll_have_md_lock(inode, &bits, lock->l_req_mode);
258 if (bits & MDS_INODELOCK_OPEN) {
261 switch (lock->l_req_mode) {
272 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
276 ll_md_real_close(inode, fmode);
278 bits &= ~MDS_INODELOCK_OPEN;
281 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
282 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
284 ll_have_md_lock(inode, &bits, LCK_MINMODE);
286 if (bits & MDS_INODELOCK_DOM) {
287 rc = ll_dom_lock_cancel(inode, lock);
289 CDEBUG(D_INODE, "cannot flush DoM data "
291 PFID(ll_inode2fid(inode)), rc);
292 lock_res_and_lock(lock);
293 ldlm_set_kms_ignore(lock);
294 unlock_res_and_lock(lock);
297 if (bits & MDS_INODELOCK_LAYOUT) {
298 struct cl_object_conf conf = {
299 .coc_opc = OBJECT_CONF_INVALIDATE,
303 rc = ll_layout_conf(inode, &conf);
305 CDEBUG(D_INODE, "cannot invalidate layout of "
307 PFID(ll_inode2fid(inode)), rc);
310 if (bits & MDS_INODELOCK_UPDATE) {
311 struct ll_inode_info *lli = ll_i2info(inode);
313 lli->lli_update_atime = 1;
316 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
317 struct ll_inode_info *lli = ll_i2info(inode);
319 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
320 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
321 lli, PFID(&lli->lli_pfid));
322 truncate_inode_pages(inode->i_mapping, 0);
324 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
325 struct inode *master_inode = NULL;
328 /* This is slave inode, since all of the child dentry
329 * is connected on the master inode, so we have to
330 * invalidate the negative children on master inode */
331 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
332 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
334 hash = cl_fid_build_ino(&lli->lli_pfid,
335 ll_need_32bit_api(ll_i2sbi(inode)));
337 /* Do not lookup the inode with ilookup5, otherwise
338 * it will cause dead lock,
339 * 1. Client1 send chmod req to the MDT0, then on MDT0,
340 * it enqueues master and all of its slaves lock,
341 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
342 * master and stripe0 lock, it will send the enqueue
343 * req (for stripe1) to MDT1, then MDT1 finds the lock
344 * has been granted to client2. Then MDT1 sends blocking
346 * 2. At the same time, client2 tries to unlink
347 * the striped dir (rm -rf striped_dir), and during
348 * lookup, it will hold the master inode of the striped
349 * directory, whose inode state is NEW, then tries to
350 * revalidate all of its slaves, (ll_prep_inode()->
351 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
352 * And it will be blocked on the server side because
354 * 3. Then the client get the blocking_ast req, cancel
355 * the lock, but being blocked if using ->ilookup5()),
356 * because master inode state is NEW. */
357 master_inode = ilookup5_nowait(inode->i_sb, hash,
358 ll_test_inode_by_fid,
359 (void *)&lli->lli_pfid);
361 ll_invalidate_negative_children(master_inode);
365 ll_invalidate_negative_children(inode);
369 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
370 inode->i_sb->s_root != NULL &&
371 inode != inode->i_sb->s_root->d_inode)
372 ll_invalidate_aliases(inode);
377 /* Check if the given lock may be downgraded instead of canceling and
378 * that convert is really needed. */
379 int ll_md_need_convert(struct ldlm_lock *lock)
381 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
383 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
384 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
385 enum ldlm_mode mode = LCK_MINMODE;
387 if (!lock->l_conn_export ||
388 !exp_connect_lock_convert(lock->l_conn_export))
391 if (!wanted || !bits || ldlm_is_cancel(lock))
394 /* do not convert locks other than DOM for now */
395 if (!((bits | wanted) & MDS_INODELOCK_DOM))
398 /* We may have already remaining bits in some other lock so
399 * lock convert will leave us just extra lock for the same bit.
400 * Check if client has other lock with the same bits and the same
401 * or lower mode and don't convert if any.
403 switch (lock->l_req_mode) {
415 /* do not convert other modes */
419 /* is lock is too old to be converted? */
420 lock_res_and_lock(lock);
421 if (ktime_after(ktime_get(),
422 ktime_add(lock->l_last_used,
423 ktime_set(ns->ns_dirty_age_limit, 0)))) {
424 unlock_res_and_lock(lock);
427 unlock_res_and_lock(lock);
429 inode = ll_inode_from_resource_lock(lock);
430 ll_have_md_lock(inode, &bits, mode);
435 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
436 void *data, int flag)
438 struct lustre_handle lockh;
439 __u64 bits = lock->l_policy_data.l_inodebits.bits;
445 case LDLM_CB_BLOCKING:
447 __u64 cancel_flags = LCF_ASYNC;
449 if (ll_md_need_convert(lock)) {
450 cancel_flags |= LCF_CONVERT;
451 /* For lock convert some cancel actions may require
452 * this lock with non-dropped canceled bits, e.g. page
453 * flush for DOM lock. So call ll_lock_cancel_bits()
454 * here while canceled bits are still set.
456 bits = lock->l_policy_data.l_inodebits.cancel_bits;
457 if (bits & MDS_INODELOCK_DOM)
458 ll_lock_cancel_bits(lock, MDS_INODELOCK_DOM);
460 ldlm_lock2handle(lock, &lockh);
461 rc = ldlm_cli_cancel(&lockh, cancel_flags);
463 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
468 case LDLM_CB_CANCELING:
469 /* Nothing to do for non-granted locks */
470 if (!ldlm_is_granted(lock))
473 if (ldlm_is_converting(lock)) {
474 /* this is called on already converted lock, so
475 * ibits has remained bits only and cancel_bits
476 * are bits that were dropped.
477 * Note that DOM lock is handled prior lock convert
478 * and is excluded here.
480 bits = lock->l_policy_data.l_inodebits.cancel_bits &
483 LASSERT(ldlm_is_canceling(lock));
485 ll_lock_cancel_bits(lock, bits);
494 __u32 ll_i2suppgid(struct inode *i)
496 if (in_group_p(i->i_gid))
497 return (__u32)from_kgid(&init_user_ns, i->i_gid);
499 return (__u32) __kgid_val(INVALID_GID);
502 /* Pack the required supplementary groups into the supplied groups array.
503 * If we don't need to use the groups from the target inode(s) then we
504 * instead pack one or more groups from the user's supplementary group
505 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
506 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
509 LASSERT(suppgids != NULL);
511 suppgids[0] = ll_i2suppgid(i1);
514 suppgids[1] = ll_i2suppgid(i2);
520 * try to reuse three types of dentry:
521 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
522 * by concurrent .revalidate).
523 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
524 * be cleared by others calling d_lustre_revalidate).
525 * 3. DISCONNECTED alias.
527 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
529 struct dentry *alias, *discon_alias, *invalid_alias;
530 DECLARE_LL_D_HLIST_NODE_PTR(p);
532 if (ll_d_hlist_empty(&inode->i_dentry))
535 discon_alias = invalid_alias = NULL;
537 ll_lock_dcache(inode);
538 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
539 LASSERT(alias != dentry);
541 spin_lock(&alias->d_lock);
542 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
543 S_ISDIR(inode->i_mode))
544 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
545 discon_alias = alias;
546 else if (alias->d_parent == dentry->d_parent &&
547 alias->d_name.hash == dentry->d_name.hash &&
548 alias->d_name.len == dentry->d_name.len &&
549 memcmp(alias->d_name.name, dentry->d_name.name,
550 dentry->d_name.len) == 0)
551 invalid_alias = alias;
552 spin_unlock(&alias->d_lock);
557 alias = invalid_alias ?: discon_alias ?: NULL;
559 spin_lock(&alias->d_lock);
561 spin_unlock(&alias->d_lock);
563 ll_unlock_dcache(inode);
569 * Similar to d_splice_alias(), but lustre treats invalid alias
570 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
572 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
578 new = ll_find_alias(inode, de);
588 "Reuse dentry %p inode %p refc %d flags %#x\n",
589 new, new->d_inode, ll_d_count(new), new->d_flags);
597 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
598 de, de->d_inode, ll_d_count(de), de->d_flags);
602 static int ll_lookup_it_finish(struct ptlrpc_request *request,
603 struct lookup_intent *it,
604 struct inode *parent, struct dentry **de,
605 void *secctx, __u32 secctxlen)
607 struct inode *inode = NULL;
610 struct dentry *alias;
613 /* NB 1 request reference will be taken away by ll_intent_lock()
615 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
617 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
618 struct req_capsule *pill = &request->rq_pill;
619 struct mdt_body *body = req_capsule_server_get(pill,
622 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
626 if (it->it_op & IT_OPEN)
627 ll_dom_finish_open(inode, request, it);
629 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
631 /* We used to query real size from OSTs here, but actually
632 * this is not needed. For stat() calls size would be updated
633 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
635 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
636 * Everybody else who needs correct file size would call
637 * ll_glimpse_size or some equivalent themselves anyway.
641 /* If security context was returned by MDT, put it in
642 * inode now to save an extra getxattr from security hooks,
643 * and avoid deadlock.
645 if (body->mbo_valid & OBD_MD_SECCTX) {
646 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
647 secctxlen = req_capsule_get_size(pill,
652 CDEBUG(D_SEC, "server returned security context"
654 PFID(ll_inode2fid(inode)));
657 if (secctx != NULL && secctxlen != 0) {
659 rc = security_inode_notifysecctx(inode, secctx,
663 CWARN("cannot set security context for "
665 PFID(ll_inode2fid(inode)), rc);
669 /* Only hash *de if it is unhashed (new dentry).
670 * Atoimc_open may passin hashed dentries for open.
672 alias = ll_splice_alias(inode, *de);
674 GOTO(out, rc = PTR_ERR(alias));
678 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
679 /* we have lookup look - unhide dentry */
680 if (bits & MDS_INODELOCK_LOOKUP)
681 d_lustre_revalidate(*de);
682 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
683 /* If file created on server, don't depend on parent UPDATE
684 * lock to unhide it. It is left hidden and next lookup can
685 * find it in ll_splice_alias.
687 /* Check that parent has UPDATE lock. */
688 struct lookup_intent parent_it = {
690 .it_lock_handle = 0 };
691 struct lu_fid fid = ll_i2info(parent)->lli_fid;
693 /* If it is striped directory, get the real stripe parent */
694 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
695 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
696 ll_i2info(parent)->lli_lsm_md,
698 (*de)->d_name.len, &fid);
703 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
705 d_lustre_revalidate(*de);
706 ll_intent_release(&parent_it);
713 if (rc != 0 && it->it_op & IT_OPEN)
714 ll_open_cleanup((*de)->d_sb, request);
719 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
720 struct lookup_intent *it,
721 void **secctx, __u32 *secctxlen)
723 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
724 struct dentry *save = dentry, *retval;
725 struct ptlrpc_request *req = NULL;
726 struct md_op_data *op_data = NULL;
729 char secctx_name[XATTR_NAME_MAX + 1];
733 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
734 RETURN(ERR_PTR(-ENAMETOOLONG));
736 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
737 dentry->d_name.len, dentry->d_name.name,
738 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
740 if (d_mountpoint(dentry))
741 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
743 if (it == NULL || it->it_op == IT_GETXATTR)
746 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
747 rc = ll_statahead(parent, &dentry, 0);
749 RETURN(dentry == save ? NULL : dentry);
752 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
753 dentry->d_sb->s_flags & MS_RDONLY)
754 RETURN(ERR_PTR(-EROFS));
756 if (it->it_op & IT_CREAT)
757 opc = LUSTRE_OPC_CREATE;
759 opc = LUSTRE_OPC_ANY;
761 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
762 dentry->d_name.len, 0, opc, NULL);
764 GOTO(out, retval = ERR_CAST(op_data));
766 /* enforce umask if acl disabled or MDS doesn't support umask */
767 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
768 it->it_create_mode &= ~current_umask();
770 if (it->it_op & IT_CREAT &&
771 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
772 rc = ll_dentry_init_security(dentry, it->it_create_mode,
774 &op_data->op_file_secctx_name,
775 &op_data->op_file_secctx,
776 &op_data->op_file_secctx_size);
778 GOTO(out, retval = ERR_PTR(rc));
780 *secctx = op_data->op_file_secctx;
781 if (secctxlen != NULL)
782 *secctxlen = op_data->op_file_secctx_size;
786 if (secctxlen != NULL)
790 /* ask for security context upon intent */
791 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
792 /* get name of security xattr to request to server */
793 rc = ll_listsecurity(parent, secctx_name,
794 sizeof(secctx_name));
796 CDEBUG(D_SEC, "cannot get security xattr name for "
798 PFID(ll_inode2fid(parent)), rc);
800 op_data->op_file_secctx_name = secctx_name;
801 op_data->op_file_secctx_name_size = rc;
802 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
803 rc, secctx_name, PFID(ll_inode2fid(parent)));
807 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
808 &ll_md_blocking_ast, 0);
809 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
810 * client does not know which suppgid should be sent to the MDS, or
811 * some other(s) changed the target file's GID after this RPC sent
812 * to the MDS with the suppgid as the original GID, then we should
813 * try again with right suppgid. */
814 if (rc == -EACCES && it->it_op & IT_OPEN &&
815 it_disposition(it, DISP_OPEN_DENY)) {
816 struct mdt_body *body;
818 LASSERT(req != NULL);
820 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
821 if (op_data->op_suppgids[0] == body->mbo_gid ||
822 op_data->op_suppgids[1] == body->mbo_gid ||
823 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
824 GOTO(out, retval = ERR_PTR(-EACCES));
826 fid_zero(&op_data->op_fid2);
827 op_data->op_suppgids[1] = body->mbo_gid;
828 ptlrpc_req_finished(req);
830 ll_intent_release(it);
831 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
832 &ll_md_blocking_ast, 0);
836 GOTO(out, retval = ERR_PTR(rc));
838 /* dir layout may change */
839 ll_unlock_md_op_lsm(op_data);
840 rc = ll_lookup_it_finish(req, it, parent, &dentry,
841 secctx != NULL ? *secctx : NULL,
842 secctxlen != NULL ? *secctxlen : 0);
844 ll_intent_release(it);
845 GOTO(out, retval = ERR_PTR(rc));
848 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
849 !S_ISREG(dentry->d_inode->i_mode) &&
850 !S_ISDIR(dentry->d_inode->i_mode)) {
851 ll_release_openhandle(dentry, it);
853 ll_lookup_finish_locks(it, dentry);
855 GOTO(out, retval = (dentry == save) ? NULL : dentry);
858 if (op_data != NULL && !IS_ERR(op_data)) {
859 if (secctx != NULL && secctxlen != NULL) {
860 /* caller needs sec ctx info, so reset it in op_data to
861 * prevent it from being freed */
862 op_data->op_file_secctx = NULL;
863 op_data->op_file_secctx_size = 0;
865 ll_finish_md_op_data(op_data);
868 ptlrpc_req_finished(req);
872 #ifdef HAVE_IOP_ATOMIC_OPEN
873 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
876 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
879 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
880 dentry->d_name.len, dentry->d_name.name,
881 PFID(ll_inode2fid(parent)), parent, flags);
884 * Optimize away (CREATE && !OPEN). Let .create handle the race.
885 * but only if we have write permissions there, otherwise we need
886 * to proceed with lookup. LU-4185
888 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
889 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
892 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
896 de = ll_lookup_it(parent, dentry, itp, NULL, NULL);
899 ll_intent_release(itp);
905 * For cached negative dentry and new dentry, handle lookup/create/open
908 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
909 struct file *file, unsigned open_flags,
910 umode_t mode, int *opened)
912 struct lookup_intent *it;
914 long long lookup_flags = LOOKUP_OPEN;
920 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
921 "open_flags %x, mode %x opened %d\n",
922 dentry->d_name.len, dentry->d_name.name,
923 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
925 /* Only negative dentries enter here */
926 LASSERT(dentry->d_inode == NULL);
928 if (!d_unhashed(dentry)) {
929 /* A valid negative dentry that just passed revalidation,
930 * there's little point to try and open it server-side,
931 * even though there's a minuscule chance it might succeed.
932 * Either way it's a valid race to just return -ENOENT here.
934 if (!(open_flags & O_CREAT))
937 /* Otherwise we just unhash it to be rehashed afresh via
938 * lookup if necessary
943 OBD_ALLOC(it, sizeof(*it));
948 if (open_flags & O_CREAT) {
949 it->it_op |= IT_CREAT;
950 lookup_flags |= LOOKUP_CREATE;
952 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
953 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
954 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
956 /* Dentry added to dcache tree in ll_lookup_it */
957 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen);
963 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
966 if (it_disposition(it, DISP_OPEN_CREATE)) {
967 /* Dentry instantiated in ll_create_it. */
968 rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
969 security_release_secctx(secctx, secctxlen);
971 /* We dget in ll_splice_alias. */
977 *opened |= FILE_CREATED;
979 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
981 if (S_ISFIFO(dentry->d_inode->i_mode)) {
982 /* We cannot call open here as it might
983 * deadlock. This case is unreachable in
984 * practice because of OBD_CONNECT_NODEVOH. */
985 rc = finish_no_open(file, de);
987 file->private_data = it;
988 rc = finish_open(file, dentry, NULL, opened);
989 /* We dget in ll_splice_alias. finish_open takes
990 * care of dget for fd open.
996 rc = finish_no_open(file, de);
1001 ll_intent_release(it);
1002 OBD_FREE(it, sizeof(*it));
1007 #else /* !HAVE_IOP_ATOMIC_OPEN */
1008 static struct lookup_intent *
1009 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1011 struct lookup_intent *it;
1015 return ERR_PTR(-ENOMEM);
1017 if (lookup_flags & LOOKUP_OPEN) {
1018 it->it_op = IT_OPEN;
1019 /* Avoid file creation for ro bind mount point(is_readonly) */
1020 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1021 it->it_op |= IT_CREAT;
1022 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1023 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1024 ~(is_readonly ? O_CREAT : 0));
1025 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1027 it->it_op = IT_GETATTR;
1033 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1034 struct nameidata *nd)
1039 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1040 struct lookup_intent *it;
1042 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1043 it = ll_d2d(dentry)->lld_it;
1044 ll_d2d(dentry)->lld_it = NULL;
1047 * Optimize away (CREATE && !OPEN). Let .create handle
1048 * the race. But only if we have write permissions
1049 * there, otherwise we need to proceed with lookup.
1052 if ((nd->flags & LOOKUP_CREATE) &&
1053 !(nd->flags & LOOKUP_OPEN) &&
1054 (inode_permission(parent,
1055 MAY_WRITE | MAY_EXEC) == 0))
1058 it = ll_convert_intent(&nd->intent.open, nd->flags,
1059 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1060 (nd->path.mnt->mnt_sb->s_flags & MS_RDONLY));
1062 RETURN((struct dentry *)it);
1065 de = ll_lookup_it(parent, dentry, it, NULL, NULL);
1068 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1069 if (dentry->d_inode &&
1070 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1071 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1072 /* We cannot call open here as it might
1073 * deadlock. This case is unreachable in
1074 * practice because of
1075 * OBD_CONNECT_NODEVOH. */
1079 nd->intent.open.file->private_data = it;
1080 filp = lookup_instantiate_filp(nd,
1086 de = (struct dentry *)filp;
1089 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1090 /* XXX This can only reliably work on assumption
1091 * that there are NO hashed negative dentries.*/
1092 ll_d2d(dentry)->lld_it = it;
1093 it = NULL; /* Will be freed in ll_create_nd */
1094 /* We absolutely depend on ll_create_nd to be
1095 * called to not leak this intent and possible
1096 * data attached to it */
1101 ll_intent_release(it);
1102 OBD_FREE(it, sizeof(*it));
1105 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL);
1110 #endif /* HAVE_IOP_ATOMIC_OPEN */
1112 /* We depend on "mode" being set with the proper file type/umask by now */
1113 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1115 struct inode *inode = NULL;
1116 struct ptlrpc_request *request = NULL;
1117 struct ll_sb_info *sbi = ll_i2sbi(dir);
1121 LASSERT(it && it->it_disposition);
1123 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1124 request = it->it_request;
1125 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1126 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1128 GOTO(out, inode = ERR_PTR(rc));
1130 /* Pause to allow for a race with concurrent access by fid */
1131 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1133 /* We asked for a lock on the directory, but were granted a
1134 * lock on the inode. Since we finally have an inode pointer,
1135 * stuff it in the lock. */
1136 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1137 PFID(ll_inode2fid(inode)), inode);
1138 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1141 ptlrpc_req_finished(request);
1146 * By the time this is called, we already have created the directory cache
1147 * entry for the new file, but it is so far negative - it has no inode.
1149 * We defer creating the OBD object(s) until open, to keep the intent and
1150 * non-intent code paths similar, and also because we do not have the MDS
1151 * inode number before calling ll_create_node() (which is needed for LOV),
1152 * so we would need to do yet another RPC to the MDS to store the LOV EA
1153 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1154 * lmm_size in datalen (the MDS still has code which will handle that).
1156 * If the create succeeds, we fill in the inode information
1157 * with d_instantiate().
1159 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1160 struct lookup_intent *it,
1161 void *secctx, __u32 secctxlen)
1163 struct inode *inode;
1167 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1168 dentry->d_name.len, dentry->d_name.name,
1169 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1171 rc = it_open_error(DISP_OPEN_CREATE, it);
1175 inode = ll_create_node(dir, it);
1177 RETURN(PTR_ERR(inode));
1179 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1182 /* must be done before d_instantiate, because it calls
1183 * security_d_instantiate, which means a getxattr if security
1184 * context is not set yet */
1185 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1186 inode_unlock(inode);
1191 d_instantiate(dentry, inode);
1193 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1194 rc = ll_inode_init_security(dentry, inode, dir);
1202 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1204 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1208 if (body->mbo_valid & OBD_MD_FLMTIME &&
1209 body->mbo_mtime > inode->i_mtime.tv_sec) {
1211 "setting fid " DFID " mtime from %lld to %llu\n",
1212 PFID(ll_inode2fid(inode)),
1213 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1214 inode->i_mtime.tv_sec = body->mbo_mtime;
1217 if (body->mbo_valid & OBD_MD_FLCTIME &&
1218 body->mbo_ctime > inode->i_ctime.tv_sec)
1219 inode->i_ctime.tv_sec = body->mbo_ctime;
1222 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1223 const char *tgt, umode_t mode, int rdev, __u32 opc)
1225 struct qstr *name = &dchild->d_name;
1226 struct ptlrpc_request *request = NULL;
1227 struct md_op_data *op_data;
1228 struct inode *inode = NULL;
1229 struct ll_sb_info *sbi = ll_i2sbi(dir);
1234 if (unlikely(tgt != NULL))
1235 tgt_len = strlen(tgt) + 1;
1238 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1239 name->len, 0, opc, NULL);
1240 if (IS_ERR(op_data))
1241 GOTO(err_exit, err = PTR_ERR(op_data));
1243 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1244 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1245 &op_data->op_file_secctx_name,
1246 &op_data->op_file_secctx,
1247 &op_data->op_file_secctx_size);
1249 GOTO(err_exit, err);
1252 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1253 from_kuid(&init_user_ns, current_fsuid()),
1254 from_kgid(&init_user_ns, current_fsgid()),
1255 cfs_curproc_cap_pack(), rdev, &request);
1256 if (err < 0 && err != -EREMOTE)
1257 GOTO(err_exit, err);
1259 /* If the client doesn't know where to create a subdirectory (or
1260 * in case of a race that sends the RPC to the wrong MDS), the
1261 * MDS will return -EREMOTE and the client will fetch the layout
1262 * of the directory, then create the directory on the right MDT. */
1263 if (unlikely(err == -EREMOTE)) {
1264 struct ll_inode_info *lli = ll_i2info(dir);
1265 struct lmv_user_md *lum;
1269 ptlrpc_req_finished(request);
1272 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1273 OBD_MD_DEFAULT_MEA);
1275 /* Update stripe_offset and retry */
1276 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1277 } else if (err2 == -ENODATA &&
1278 lli->lli_def_stripe_offset != -1) {
1279 /* If there are no default stripe EA on the MDT, but the
1280 * client has default stripe, then it probably means
1281 * default stripe EA has just been deleted. */
1282 lli->lli_def_stripe_offset = -1;
1284 GOTO(err_exit, err);
1287 ptlrpc_req_finished(request);
1289 ll_finish_md_op_data(op_data);
1293 ll_update_times(request, dir);
1295 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1297 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1299 GOTO(err_exit, err);
1301 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1303 /* must be done before d_instantiate, because it calls
1304 * security_d_instantiate, which means a getxattr if security
1305 * context is not set yet */
1306 err = security_inode_notifysecctx(inode,
1307 op_data->op_file_secctx,
1308 op_data->op_file_secctx_size);
1309 inode_unlock(inode);
1311 GOTO(err_exit, err);
1314 d_instantiate(dchild, inode);
1316 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1317 err = ll_inode_init_security(dchild, inode, dir);
1319 GOTO(err_exit, err);
1324 if (request != NULL)
1325 ptlrpc_req_finished(request);
1327 if (!IS_ERR_OR_NULL(op_data))
1328 ll_finish_md_op_data(op_data);
1333 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1336 struct qstr *name = &dchild->d_name;
1340 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1341 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1344 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1345 mode &= ~current_umask();
1347 switch (mode & S_IFMT) {
1349 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1355 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1366 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1371 #ifdef HAVE_IOP_ATOMIC_OPEN
1373 * Plain create. Intent create is handled in atomic_open.
1375 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1376 umode_t mode, bool want_excl)
1380 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1382 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1383 "flags=%u, excl=%d\n", dentry->d_name.len,
1384 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1385 dir, mode, want_excl);
1387 /* Using mknod(2) to create a regular file is designed to not recognize
1388 * volatile file name, so we use ll_mknod() here. */
1389 rc = ll_mknod(dir, dentry, mode, 0);
1391 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1393 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1394 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1398 #else /* !HAVE_IOP_ATOMIC_OPEN */
1399 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1400 ll_umode_t mode, struct nameidata *nd)
1402 struct ll_dentry_data *lld = ll_d2d(dentry);
1403 struct lookup_intent *it = NULL;
1406 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1412 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1413 * so that volatile file name is recoginized.
1414 * Mknod(2), however, is designed to not recognize volatile
1415 * file name to avoid inode leak under orphan directory until
1417 return ll_new_node(dir, dentry, NULL, mode, 0,
1423 /* Was there an error? Propagate it! */
1424 if (it->it_status) {
1429 rc = ll_create_it(dir, dentry, it, NULL, 0);
1430 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1433 nd->intent.open.file->private_data = it;
1434 filp = lookup_instantiate_filp(nd, dentry, NULL);
1440 ll_intent_release(it);
1441 OBD_FREE(it, sizeof(*it));
1444 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1448 #endif /* HAVE_IOP_ATOMIC_OPEN */
1450 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1451 const char *oldpath)
1453 struct qstr *name = &dchild->d_name;
1457 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1458 name->len, name->name, PFID(ll_inode2fid(dir)),
1459 dir, 3000, oldpath);
1461 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1462 LUSTRE_OPC_SYMLINK);
1465 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1470 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1471 struct dentry *new_dentry)
1473 struct inode *src = old_dentry->d_inode;
1474 struct qstr *name = &new_dentry->d_name;
1475 struct ll_sb_info *sbi = ll_i2sbi(dir);
1476 struct ptlrpc_request *request = NULL;
1477 struct md_op_data *op_data;
1481 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1482 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1483 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1485 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1486 0, LUSTRE_OPC_ANY, NULL);
1487 if (IS_ERR(op_data))
1488 RETURN(PTR_ERR(op_data));
1490 err = md_link(sbi->ll_md_exp, op_data, &request);
1491 ll_finish_md_op_data(op_data);
1495 ll_update_times(request, dir);
1496 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1499 ptlrpc_req_finished(request);
1503 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1505 struct qstr *name = &dchild->d_name;
1509 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1510 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1512 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1513 mode &= ~current_umask();
1515 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1517 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1519 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1524 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1526 struct qstr *name = &dchild->d_name;
1527 struct ptlrpc_request *request = NULL;
1528 struct md_op_data *op_data;
1532 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1533 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1535 if (unlikely(d_mountpoint(dchild)))
1538 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1539 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1540 if (IS_ERR(op_data))
1541 RETURN(PTR_ERR(op_data));
1543 if (dchild->d_inode != NULL)
1544 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1546 op_data->op_fid2 = op_data->op_fid3;
1547 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1548 ll_finish_md_op_data(op_data);
1550 ll_update_times(request, dir);
1551 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1554 ptlrpc_req_finished(request);
1561 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1563 struct ptlrpc_request *request = NULL;
1564 struct md_op_data *op_data;
1568 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1569 namelen, name, PFID(ll_inode2fid(dir)), dir);
1571 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1572 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1573 if (IS_ERR(op_data))
1574 RETURN(PTR_ERR(op_data));
1575 op_data->op_cli_flags |= CLI_RM_ENTRY;
1576 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1577 ll_finish_md_op_data(op_data);
1579 ll_update_times(request, dir);
1580 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1583 ptlrpc_req_finished(request);
1587 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1589 struct qstr *name = &dchild->d_name;
1590 struct ptlrpc_request *request = NULL;
1591 struct md_op_data *op_data;
1592 struct mdt_body *body;
1595 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1596 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1599 * XXX: unlink bind mountpoint maybe call to here,
1600 * just check it as vfs_unlink does.
1602 if (unlikely(d_mountpoint(dchild)))
1605 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1606 LUSTRE_OPC_ANY, NULL);
1607 if (IS_ERR(op_data))
1608 RETURN(PTR_ERR(op_data));
1610 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1612 op_data->op_fid2 = op_data->op_fid3;
1613 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1614 ll_finish_md_op_data(op_data);
1619 * The server puts attributes in on the last unlink, use them to update
1620 * the link count so the inode can be freed immediately.
1622 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1623 if (body->mbo_valid & OBD_MD_FLNLINK)
1624 set_nlink(dchild->d_inode, body->mbo_nlink);
1626 ll_update_times(request, dir);
1627 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1630 ptlrpc_req_finished(request);
1634 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1635 struct inode *tgt, struct dentry *tgt_dchild
1636 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1637 , unsigned int flags
1641 struct qstr *src_name = &src_dchild->d_name;
1642 struct qstr *tgt_name = &tgt_dchild->d_name;
1643 struct ptlrpc_request *request = NULL;
1644 struct ll_sb_info *sbi = ll_i2sbi(src);
1645 struct md_op_data *op_data;
1649 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1654 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1655 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1656 src_name->len, src_name->name,
1657 PFID(ll_inode2fid(src)), src, tgt_name->len,
1658 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1660 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1663 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1664 LUSTRE_OPC_ANY, NULL);
1665 if (IS_ERR(op_data))
1666 RETURN(PTR_ERR(op_data));
1668 if (src_dchild->d_inode != NULL)
1669 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1671 if (tgt_dchild->d_inode != NULL)
1672 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1674 err = md_rename(sbi->ll_md_exp, op_data,
1675 src_name->name, src_name->len,
1676 tgt_name->name, tgt_name->len, &request);
1677 ll_finish_md_op_data(op_data);
1679 ll_update_times(request, src);
1680 ll_update_times(request, tgt);
1681 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1684 ptlrpc_req_finished(request);
1687 d_move(src_dchild, tgt_dchild);
1692 const struct inode_operations ll_dir_inode_operations = {
1694 #ifdef HAVE_IOP_ATOMIC_OPEN
1695 .atomic_open = ll_atomic_open,
1697 .lookup = ll_lookup_nd,
1698 .create = ll_create_nd,
1699 /* We need all these non-raw things for NFSD, to not patch it. */
1700 .unlink = ll_unlink,
1703 .symlink = ll_symlink,
1705 .rename = ll_rename,
1706 .setattr = ll_setattr,
1707 .getattr = ll_getattr,
1708 .permission = ll_inode_permission,
1709 #ifdef HAVE_IOP_XATTR
1710 .setxattr = ll_setxattr,
1711 .getxattr = ll_getxattr,
1712 .removexattr = ll_removexattr,
1714 .listxattr = ll_listxattr,
1715 #ifdef HAVE_IOP_GET_ACL
1716 .get_acl = ll_get_acl,
1718 #ifdef HAVE_IOP_SET_ACL
1719 .set_acl = ll_set_acl,
1723 const struct inode_operations ll_special_inode_operations = {
1724 .setattr = ll_setattr,
1725 .getattr = ll_getattr,
1726 .permission = ll_inode_permission,
1727 #ifdef HAVE_IOP_XATTR
1728 .setxattr = ll_setxattr,
1729 .getxattr = ll_getxattr,
1730 .removexattr = ll_removexattr,
1732 .listxattr = ll_listxattr,
1733 #ifdef HAVE_IOP_GET_ACL
1734 .get_acl = ll_get_acl,
1736 #ifdef HAVE_IOP_SET_ACL
1737 .set_acl = ll_set_acl,