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 struct ll_inode_info *lli;
234 __u64 bits = to_cancel;
240 /* That means the inode is evicted most likely and may cause
241 * the skipping of lock cleanups below, so print the message
244 if (lock->l_resource->lr_lvb_inode)
246 "can't take inode for the lock (%sevicted)\n",
247 lock->l_resource->lr_lvb_inode->i_state &
248 I_FREEING ? "" : "not ");
252 if (!fid_res_name_eq(ll_inode2fid(inode),
253 &lock->l_resource->lr_name)) {
254 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
255 PFID(ll_inode2fid(inode)), inode);
259 if (bits & MDS_INODELOCK_XATTR) {
260 ll_xattr_cache_destroy(inode);
261 bits &= ~MDS_INODELOCK_XATTR;
264 /* For OPEN locks we differentiate between lock modes
265 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
266 if (bits & MDS_INODELOCK_OPEN)
267 ll_have_md_lock(inode, &bits, lock->l_req_mode);
269 if (bits & MDS_INODELOCK_OPEN) {
272 switch (lock->l_req_mode) {
283 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
287 ll_md_real_close(inode, fmode);
289 bits &= ~MDS_INODELOCK_OPEN;
292 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
293 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
295 ll_have_md_lock(inode, &bits, LCK_MINMODE);
297 if (bits & MDS_INODELOCK_DOM) {
298 rc = ll_dom_lock_cancel(inode, lock);
300 CDEBUG(D_INODE, "cannot flush DoM data "
302 PFID(ll_inode2fid(inode)), rc);
303 lock_res_and_lock(lock);
304 ldlm_set_kms_ignore(lock);
305 unlock_res_and_lock(lock);
308 if (bits & MDS_INODELOCK_LAYOUT) {
309 struct cl_object_conf conf = {
310 .coc_opc = OBJECT_CONF_INVALIDATE,
314 rc = ll_layout_conf(inode, &conf);
316 CDEBUG(D_INODE, "cannot invalidate layout of "
318 PFID(ll_inode2fid(inode)), rc);
321 lli = ll_i2info(inode);
323 if (bits & MDS_INODELOCK_UPDATE)
324 lli->lli_update_atime = 1;
326 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
327 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
328 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
329 lli, PFID(&lli->lli_pfid));
330 truncate_inode_pages(inode->i_mapping, 0);
332 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
333 struct inode *master_inode = NULL;
336 /* This is slave inode, since all of the child dentry
337 * is connected on the master inode, so we have to
338 * invalidate the negative children on master inode */
339 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
340 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
342 hash = cl_fid_build_ino(&lli->lli_pfid,
343 ll_need_32bit_api(ll_i2sbi(inode)));
345 /* Do not lookup the inode with ilookup5, otherwise
346 * it will cause dead lock,
347 * 1. Client1 send chmod req to the MDT0, then on MDT0,
348 * it enqueues master and all of its slaves lock,
349 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
350 * master and stripe0 lock, it will send the enqueue
351 * req (for stripe1) to MDT1, then MDT1 finds the lock
352 * has been granted to client2. Then MDT1 sends blocking
354 * 2. At the same time, client2 tries to unlink
355 * the striped dir (rm -rf striped_dir), and during
356 * lookup, it will hold the master inode of the striped
357 * directory, whose inode state is NEW, then tries to
358 * revalidate all of its slaves, (ll_prep_inode()->
359 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
360 * And it will be blocked on the server side because
362 * 3. Then the client get the blocking_ast req, cancel
363 * the lock, but being blocked if using ->ilookup5()),
364 * because master inode state is NEW. */
365 master_inode = ilookup5_nowait(inode->i_sb, hash,
366 ll_test_inode_by_fid,
367 (void *)&lli->lli_pfid);
369 ll_invalidate_negative_children(master_inode);
373 ll_invalidate_negative_children(inode);
377 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
378 inode->i_sb->s_root != NULL &&
379 inode != inode->i_sb->s_root->d_inode)
380 ll_invalidate_aliases(inode);
386 /* Check if the given lock may be downgraded instead of canceling and
387 * that convert is really needed. */
388 int ll_md_need_convert(struct ldlm_lock *lock)
390 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
392 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
393 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
394 enum ldlm_mode mode = LCK_MINMODE;
396 if (!lock->l_conn_export ||
397 !exp_connect_lock_convert(lock->l_conn_export))
400 if (!wanted || !bits || ldlm_is_cancel(lock))
403 /* do not convert locks other than DOM for now */
404 if (!((bits | wanted) & MDS_INODELOCK_DOM))
407 /* We may have already remaining bits in some other lock so
408 * lock convert will leave us just extra lock for the same bit.
409 * Check if client has other lock with the same bits and the same
410 * or lower mode and don't convert if any.
412 switch (lock->l_req_mode) {
424 /* do not convert other modes */
428 /* is lock is too old to be converted? */
429 lock_res_and_lock(lock);
430 if (ktime_after(ktime_get(),
431 ktime_add(lock->l_last_used,
432 ktime_set(ns->ns_dirty_age_limit, 0)))) {
433 unlock_res_and_lock(lock);
436 unlock_res_and_lock(lock);
438 inode = ll_inode_from_resource_lock(lock);
439 ll_have_md_lock(inode, &bits, mode);
444 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
445 void *data, int flag)
447 struct lustre_handle lockh;
448 __u64 bits = lock->l_policy_data.l_inodebits.bits;
454 case LDLM_CB_BLOCKING:
456 __u64 cancel_flags = LCF_ASYNC;
458 if (ll_md_need_convert(lock)) {
459 cancel_flags |= LCF_CONVERT;
460 /* For lock convert some cancel actions may require
461 * this lock with non-dropped canceled bits, e.g. page
462 * flush for DOM lock. So call ll_lock_cancel_bits()
463 * here while canceled bits are still set.
465 bits = lock->l_policy_data.l_inodebits.cancel_bits;
466 if (bits & MDS_INODELOCK_DOM)
467 ll_lock_cancel_bits(lock, MDS_INODELOCK_DOM);
469 ldlm_lock2handle(lock, &lockh);
470 rc = ldlm_cli_cancel(&lockh, cancel_flags);
472 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
477 case LDLM_CB_CANCELING:
478 /* Nothing to do for non-granted locks */
479 if (!ldlm_is_granted(lock))
482 if (ldlm_is_converting(lock)) {
483 /* this is called on already converted lock, so
484 * ibits has remained bits only and cancel_bits
485 * are bits that were dropped.
486 * Note that DOM lock is handled prior lock convert
487 * and is excluded here.
489 bits = lock->l_policy_data.l_inodebits.cancel_bits &
492 LASSERT(ldlm_is_canceling(lock));
494 ll_lock_cancel_bits(lock, bits);
503 __u32 ll_i2suppgid(struct inode *i)
505 if (in_group_p(i->i_gid))
506 return (__u32)from_kgid(&init_user_ns, i->i_gid);
508 return (__u32) __kgid_val(INVALID_GID);
511 /* Pack the required supplementary groups into the supplied groups array.
512 * If we don't need to use the groups from the target inode(s) then we
513 * instead pack one or more groups from the user's supplementary group
514 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
515 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
518 LASSERT(suppgids != NULL);
520 suppgids[0] = ll_i2suppgid(i1);
523 suppgids[1] = ll_i2suppgid(i2);
529 * try to reuse three types of dentry:
530 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
531 * by concurrent .revalidate).
532 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
533 * be cleared by others calling d_lustre_revalidate).
534 * 3. DISCONNECTED alias.
536 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
538 struct dentry *alias, *discon_alias, *invalid_alias;
539 DECLARE_LL_D_HLIST_NODE_PTR(p);
541 if (ll_d_hlist_empty(&inode->i_dentry))
544 discon_alias = invalid_alias = NULL;
546 ll_lock_dcache(inode);
547 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
548 LASSERT(alias != dentry);
550 spin_lock(&alias->d_lock);
551 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
552 S_ISDIR(inode->i_mode))
553 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
554 discon_alias = alias;
555 else if (alias->d_parent == dentry->d_parent &&
556 alias->d_name.hash == dentry->d_name.hash &&
557 alias->d_name.len == dentry->d_name.len &&
558 memcmp(alias->d_name.name, dentry->d_name.name,
559 dentry->d_name.len) == 0)
560 invalid_alias = alias;
561 spin_unlock(&alias->d_lock);
566 alias = invalid_alias ?: discon_alias ?: NULL;
568 spin_lock(&alias->d_lock);
570 spin_unlock(&alias->d_lock);
572 ll_unlock_dcache(inode);
578 * Similar to d_splice_alias(), but lustre treats invalid alias
579 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
581 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
587 new = ll_find_alias(inode, de);
597 "Reuse dentry %p inode %p refc %d flags %#x\n",
598 new, new->d_inode, ll_d_count(new), new->d_flags);
606 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
607 de, de->d_inode, ll_d_count(de), de->d_flags);
611 static int ll_lookup_it_finish(struct ptlrpc_request *request,
612 struct lookup_intent *it,
613 struct inode *parent, struct dentry **de,
614 void *secctx, __u32 secctxlen)
616 struct inode *inode = NULL;
619 struct dentry *alias;
622 /* NB 1 request reference will be taken away by ll_intent_lock()
624 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
626 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
627 struct req_capsule *pill = &request->rq_pill;
628 struct mdt_body *body = req_capsule_server_get(pill,
631 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
635 if (it->it_op & IT_OPEN)
636 ll_dom_finish_open(inode, request, it);
638 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
640 /* We used to query real size from OSTs here, but actually
641 * this is not needed. For stat() calls size would be updated
642 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
644 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
645 * Everybody else who needs correct file size would call
646 * ll_glimpse_size or some equivalent themselves anyway.
650 /* If security context was returned by MDT, put it in
651 * inode now to save an extra getxattr from security hooks,
652 * and avoid deadlock.
654 if (body->mbo_valid & OBD_MD_SECCTX) {
655 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
656 secctxlen = req_capsule_get_size(pill,
661 CDEBUG(D_SEC, "server returned security context"
663 PFID(ll_inode2fid(inode)));
666 if (secctx != NULL && secctxlen != 0) {
668 rc = security_inode_notifysecctx(inode, secctx,
672 CWARN("cannot set security context for "
674 PFID(ll_inode2fid(inode)), rc);
678 /* Only hash *de if it is unhashed (new dentry).
679 * Atoimc_open may passin hashed dentries for open.
681 alias = ll_splice_alias(inode, *de);
683 GOTO(out, rc = PTR_ERR(alias));
687 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
688 /* we have lookup look - unhide dentry */
689 if (bits & MDS_INODELOCK_LOOKUP)
690 d_lustre_revalidate(*de);
691 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
693 * If file was created on the server, the dentry is revalidated
694 * in ll_create_it if the lock allows for it.
696 /* Check that parent has UPDATE lock. */
697 struct lookup_intent parent_it = {
699 .it_lock_handle = 0 };
700 struct lu_fid fid = ll_i2info(parent)->lli_fid;
702 /* If it is striped directory, get the real stripe parent */
703 if (unlikely(ll_dir_striped(parent))) {
704 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
705 ll_i2info(parent)->lli_lsm_md,
707 (*de)->d_name.len, &fid);
712 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
714 d_lustre_revalidate(*de);
715 ll_intent_release(&parent_it);
722 if (rc != 0 && it->it_op & IT_OPEN)
723 ll_open_cleanup((*de)->d_sb, request);
728 struct pcc_create_attach {
729 struct pcc_dataset *pca_dataset;
730 struct dentry *pca_dentry;
733 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
734 struct lookup_intent *it,
735 void **secctx, __u32 *secctxlen,
736 struct pcc_create_attach *pca)
738 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
739 struct dentry *save = dentry, *retval;
740 struct ptlrpc_request *req = NULL;
741 struct md_op_data *op_data = NULL;
742 struct lov_user_md *lum = NULL;
745 char secctx_name[XATTR_NAME_MAX + 1];
749 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
750 RETURN(ERR_PTR(-ENAMETOOLONG));
752 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
753 dentry->d_name.len, dentry->d_name.name,
754 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
756 if (d_mountpoint(dentry))
757 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
759 if (it == NULL || it->it_op == IT_GETXATTR)
762 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
763 rc = ll_statahead(parent, &dentry, 0);
765 RETURN(dentry == save ? NULL : dentry);
768 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
769 dentry->d_sb->s_flags & MS_RDONLY)
770 RETURN(ERR_PTR(-EROFS));
772 if (it->it_op & IT_CREAT)
773 opc = LUSTRE_OPC_CREATE;
775 opc = LUSTRE_OPC_ANY;
777 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
778 dentry->d_name.len, 0, opc, NULL);
780 GOTO(out, retval = ERR_CAST(op_data));
782 /* enforce umask if acl disabled or MDS doesn't support umask */
783 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
784 it->it_create_mode &= ~current_umask();
786 if (it->it_op & IT_CREAT &&
787 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
788 rc = ll_dentry_init_security(dentry, it->it_create_mode,
790 &op_data->op_file_secctx_name,
791 &op_data->op_file_secctx,
792 &op_data->op_file_secctx_size);
794 GOTO(out, retval = ERR_PTR(rc));
796 *secctx = op_data->op_file_secctx;
797 if (secctxlen != NULL)
798 *secctxlen = op_data->op_file_secctx_size;
802 if (secctxlen != NULL)
806 /* ask for security context upon intent */
807 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
808 /* get name of security xattr to request to server */
809 rc = ll_listsecurity(parent, secctx_name,
810 sizeof(secctx_name));
812 CDEBUG(D_SEC, "cannot get security xattr name for "
814 PFID(ll_inode2fid(parent)), rc);
816 op_data->op_file_secctx_name = secctx_name;
817 op_data->op_file_secctx_name_size = rc;
818 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
819 rc, secctx_name, PFID(ll_inode2fid(parent)));
823 if (pca && pca->pca_dataset) {
824 struct pcc_dataset *dataset = pca->pca_dataset;
828 GOTO(out, retval = ERR_PTR(-ENOMEM));
830 lum->lmm_magic = LOV_USER_MAGIC_V1;
831 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
832 op_data->op_data = lum;
833 op_data->op_data_size = sizeof(*lum);
834 op_data->op_archive_id = dataset->pccd_rwid;
836 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
839 GOTO(out, retval = ERR_PTR(rc));
841 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
844 GOTO(out, retval = ERR_PTR(rc));
846 it->it_flags |= MDS_OPEN_PCC;
849 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
850 &ll_md_blocking_ast, 0);
851 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
852 * client does not know which suppgid should be sent to the MDS, or
853 * some other(s) changed the target file's GID after this RPC sent
854 * to the MDS with the suppgid as the original GID, then we should
855 * try again with right suppgid. */
856 if (rc == -EACCES && it->it_op & IT_OPEN &&
857 it_disposition(it, DISP_OPEN_DENY)) {
858 struct mdt_body *body;
860 LASSERT(req != NULL);
862 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
863 if (op_data->op_suppgids[0] == body->mbo_gid ||
864 op_data->op_suppgids[1] == body->mbo_gid ||
865 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
866 GOTO(out, retval = ERR_PTR(-EACCES));
868 fid_zero(&op_data->op_fid2);
869 op_data->op_suppgids[1] = body->mbo_gid;
870 ptlrpc_req_finished(req);
872 ll_intent_release(it);
873 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
874 &ll_md_blocking_ast, 0);
878 GOTO(out, retval = ERR_PTR(rc));
880 /* dir layout may change */
881 ll_unlock_md_op_lsm(op_data);
882 rc = ll_lookup_it_finish(req, it, parent, &dentry,
883 secctx != NULL ? *secctx : NULL,
884 secctxlen != NULL ? *secctxlen : 0);
886 ll_intent_release(it);
887 GOTO(out, retval = ERR_PTR(rc));
890 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
891 !S_ISREG(dentry->d_inode->i_mode) &&
892 !S_ISDIR(dentry->d_inode->i_mode)) {
893 ll_release_openhandle(dentry, it);
895 ll_lookup_finish_locks(it, dentry);
897 GOTO(out, retval = (dentry == save) ? NULL : dentry);
900 if (op_data != NULL && !IS_ERR(op_data)) {
901 if (secctx != NULL && secctxlen != NULL) {
902 /* caller needs sec ctx info, so reset it in op_data to
903 * prevent it from being freed */
904 op_data->op_file_secctx = NULL;
905 op_data->op_file_secctx_size = 0;
907 ll_finish_md_op_data(op_data);
913 ptlrpc_req_finished(req);
917 #ifdef HAVE_IOP_ATOMIC_OPEN
918 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
921 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
924 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
925 dentry->d_name.len, dentry->d_name.name,
926 PFID(ll_inode2fid(parent)), parent, flags);
929 * Optimize away (CREATE && !OPEN). Let .create handle the race.
930 * but only if we have write permissions there, otherwise we need
931 * to proceed with lookup. LU-4185
933 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
934 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
937 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
941 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL);
944 ll_intent_release(itp);
949 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
950 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
951 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
953 # define ll_set_created(o, f) \
955 (f)->f_mode |= FMODE_CREATED; \
959 # define ll_is_opened(o, f) (*(o))
960 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
961 # define ll_last_arg , int *opened
962 # define ll_set_created(o, f) \
964 *(o) |= FILE_CREATED; \
970 * For cached negative dentry and new dentry, handle lookup/create/open
973 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
974 struct file *file, unsigned open_flags,
975 umode_t mode ll_last_arg)
977 struct lookup_intent *it;
979 long long lookup_flags = LOOKUP_OPEN;
982 struct ll_sb_info *sbi;
983 struct pcc_create_attach pca = {NULL, NULL};
984 struct pcc_dataset *dataset = NULL;
988 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
989 "open_flags %x, mode %x opened %d\n",
990 dentry->d_name.len, dentry->d_name.name,
991 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
992 ll_is_opened(opened, file));
994 /* Only negative dentries enter here */
995 LASSERT(dentry->d_inode == NULL);
997 if (!d_unhashed(dentry)) {
998 /* A valid negative dentry that just passed revalidation,
999 * there's little point to try and open it server-side,
1000 * even though there's a minuscule chance it might succeed.
1001 * Either way it's a valid race to just return -ENOENT here.
1003 if (!(open_flags & O_CREAT))
1006 /* Otherwise we just unhash it to be rehashed afresh via
1007 * lookup if necessary
1012 OBD_ALLOC(it, sizeof(*it));
1016 it->it_op = IT_OPEN;
1017 if (open_flags & O_CREAT) {
1018 it->it_op |= IT_CREAT;
1019 lookup_flags |= LOOKUP_CREATE;
1020 sbi = ll_i2sbi(dir);
1021 /* Volatile file is used for HSM restore, so do not use PCC */
1022 if (!filename_is_volatile(dentry->d_name.name,
1023 dentry->d_name.len, NULL)) {
1024 struct pcc_matcher item;
1026 item.pm_uid = from_kuid(&init_user_ns, current_uid());
1027 item.pm_gid = from_kgid(&init_user_ns, current_gid());
1028 item.pm_projid = ll_i2info(dir)->lli_projid;
1029 item.pm_name = &dentry->d_name;
1030 dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1032 pca.pca_dataset = dataset;
1035 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1036 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1037 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1039 /* Dentry added to dcache tree in ll_lookup_it */
1040 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca);
1043 else if (de != NULL)
1046 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1049 if (it_disposition(it, DISP_OPEN_CREATE)) {
1050 /* Dentry instantiated in ll_create_it. */
1051 rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
1052 security_release_secctx(secctx, secctxlen);
1054 /* We dget in ll_splice_alias. */
1059 if (dataset != NULL && dentry->d_inode) {
1060 rc = pcc_inode_create_fini(dataset,
1066 GOTO(out_release, rc);
1069 ll_set_created(opened, file);
1072 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1074 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1075 /* We cannot call open here as it might
1076 * deadlock. This case is unreachable in
1077 * practice because of OBD_CONNECT_NODEVOH. */
1078 rc = finish_no_open(file, de);
1080 file->private_data = it;
1081 rc = ll_finish_open(file, dentry, opened);
1082 /* We dget in ll_splice_alias. finish_open takes
1083 * care of dget for fd open.
1089 rc = finish_no_open(file, de);
1094 if (dataset != NULL)
1095 pcc_dataset_put(dataset);
1096 ll_intent_release(it);
1097 OBD_FREE(it, sizeof(*it));
1102 #else /* !HAVE_IOP_ATOMIC_OPEN */
1103 static struct lookup_intent *
1104 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1106 struct lookup_intent *it;
1110 return ERR_PTR(-ENOMEM);
1112 if (lookup_flags & LOOKUP_OPEN) {
1113 it->it_op = IT_OPEN;
1114 /* Avoid file creation for ro bind mount point(is_readonly) */
1115 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1116 it->it_op |= IT_CREAT;
1117 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1118 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1119 ~(is_readonly ? O_CREAT : 0));
1120 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1122 it->it_op = IT_GETATTR;
1128 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1129 struct nameidata *nd)
1134 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1135 struct lookup_intent *it;
1137 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1138 it = ll_d2d(dentry)->lld_it;
1139 ll_d2d(dentry)->lld_it = NULL;
1142 * Optimize away (CREATE && !OPEN). Let .create handle
1143 * the race. But only if we have write permissions
1144 * there, otherwise we need to proceed with lookup.
1147 if ((nd->flags & LOOKUP_CREATE) &&
1148 !(nd->flags & LOOKUP_OPEN) &&
1149 (inode_permission(parent,
1150 MAY_WRITE | MAY_EXEC) == 0))
1153 it = ll_convert_intent(&nd->intent.open, nd->flags,
1154 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1155 (nd->path.mnt->mnt_sb->s_flags & MS_RDONLY));
1157 RETURN((struct dentry *)it);
1160 de = ll_lookup_it(parent, dentry, it, NULL, NULL, NULL);
1163 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1164 if (dentry->d_inode &&
1165 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1166 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1167 /* We cannot call open here as it might
1168 * deadlock. This case is unreachable in
1169 * practice because of
1170 * OBD_CONNECT_NODEVOH. */
1174 nd->intent.open.file->private_data = it;
1175 filp = lookup_instantiate_filp(nd,
1181 de = (struct dentry *)filp;
1184 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1185 /* XXX This can only reliably work on assumption
1186 * that there are NO hashed negative dentries.*/
1187 ll_d2d(dentry)->lld_it = it;
1188 it = NULL; /* Will be freed in ll_create_nd */
1189 /* We absolutely depend on ll_create_nd to be
1190 * called to not leak this intent and possible
1191 * data attached to it */
1196 ll_intent_release(it);
1197 OBD_FREE(it, sizeof(*it));
1200 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL, NULL);
1205 #endif /* HAVE_IOP_ATOMIC_OPEN */
1207 /* We depend on "mode" being set with the proper file type/umask by now */
1208 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1210 struct inode *inode = NULL;
1211 struct ptlrpc_request *request = NULL;
1212 struct ll_sb_info *sbi = ll_i2sbi(dir);
1216 LASSERT(it && it->it_disposition);
1218 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1219 request = it->it_request;
1220 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1221 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1223 GOTO(out, inode = ERR_PTR(rc));
1225 /* Pause to allow for a race with concurrent access by fid */
1226 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1228 /* We asked for a lock on the directory, but were granted a
1229 * lock on the inode. Since we finally have an inode pointer,
1230 * stuff it in the lock. */
1231 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1232 PFID(ll_inode2fid(inode)), inode);
1233 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1236 ptlrpc_req_finished(request);
1241 * By the time this is called, we already have created the directory cache
1242 * entry for the new file, but it is so far negative - it has no inode.
1244 * We defer creating the OBD object(s) until open, to keep the intent and
1245 * non-intent code paths similar, and also because we do not have the MDS
1246 * inode number before calling ll_create_node() (which is needed for LOV),
1247 * so we would need to do yet another RPC to the MDS to store the LOV EA
1248 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1249 * lmm_size in datalen (the MDS still has code which will handle that).
1251 * If the create succeeds, we fill in the inode information
1252 * with d_instantiate().
1254 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1255 struct lookup_intent *it,
1256 void *secctx, __u32 secctxlen)
1258 struct inode *inode;
1263 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1264 dentry->d_name.len, dentry->d_name.name,
1265 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1267 rc = it_open_error(DISP_OPEN_CREATE, it);
1271 inode = ll_create_node(dir, it);
1273 RETURN(PTR_ERR(inode));
1275 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1278 /* must be done before d_instantiate, because it calls
1279 * security_d_instantiate, which means a getxattr if security
1280 * context is not set yet */
1281 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1282 inode_unlock(inode);
1287 d_instantiate(dentry, inode);
1289 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1290 rc = ll_inode_init_security(dentry, inode, dir);
1295 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1296 if (bits & MDS_INODELOCK_LOOKUP)
1297 d_lustre_revalidate(dentry);
1302 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1304 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1308 if (body->mbo_valid & OBD_MD_FLMTIME &&
1309 body->mbo_mtime > inode->i_mtime.tv_sec) {
1311 "setting fid " DFID " mtime from %lld to %llu\n",
1312 PFID(ll_inode2fid(inode)),
1313 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1314 inode->i_mtime.tv_sec = body->mbo_mtime;
1317 if (body->mbo_valid & OBD_MD_FLCTIME &&
1318 body->mbo_ctime > inode->i_ctime.tv_sec)
1319 inode->i_ctime.tv_sec = body->mbo_ctime;
1322 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1323 const char *tgt, umode_t mode, int rdev, __u32 opc)
1325 struct qstr *name = &dchild->d_name;
1326 struct ptlrpc_request *request = NULL;
1327 struct md_op_data *op_data;
1328 struct inode *inode = NULL;
1329 struct ll_sb_info *sbi = ll_i2sbi(dir);
1334 if (unlikely(tgt != NULL))
1335 tgt_len = strlen(tgt) + 1;
1338 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1339 name->len, 0, opc, NULL);
1340 if (IS_ERR(op_data))
1341 GOTO(err_exit, err = PTR_ERR(op_data));
1343 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1344 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1345 &op_data->op_file_secctx_name,
1346 &op_data->op_file_secctx,
1347 &op_data->op_file_secctx_size);
1349 GOTO(err_exit, err);
1352 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1353 from_kuid(&init_user_ns, current_fsuid()),
1354 from_kgid(&init_user_ns, current_fsgid()),
1355 cfs_curproc_cap_pack(), rdev, &request);
1356 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1358 * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1359 * fetch default LMV here.
1361 if (unlikely(err == -EREMOTE)) {
1362 struct ll_inode_info *lli = ll_i2info(dir);
1363 struct lmv_user_md *lum;
1367 ptlrpc_req_finished(request);
1370 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1371 OBD_MD_DEFAULT_MEA);
1372 ll_finish_md_op_data(op_data);
1375 struct lustre_md md = { NULL };
1377 md.body = req_capsule_server_get(&request->rq_pill,
1380 GOTO(err_exit, err = -EPROTO);
1382 OBD_ALLOC_PTR(md.default_lmv);
1383 if (!md.default_lmv)
1384 GOTO(err_exit, err = -ENOMEM);
1386 md.default_lmv->lsm_md_magic = lum->lum_magic;
1387 md.default_lmv->lsm_md_stripe_count =
1388 lum->lum_stripe_count;
1389 md.default_lmv->lsm_md_master_mdt_index =
1390 lum->lum_stripe_offset;
1391 md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1393 err = ll_update_inode(dir, &md);
1394 md_free_lustre_md(sbi->ll_md_exp, &md);
1396 GOTO(err_exit, err);
1397 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1399 * If there are no default stripe EA on the MDT, but the
1400 * client has default stripe, then it probably means
1401 * default stripe EA has just been deleted.
1403 down_write(&lli->lli_lsm_sem);
1404 if (lli->lli_default_lsm_md)
1405 OBD_FREE_PTR(lli->lli_default_lsm_md);
1406 lli->lli_default_lsm_md = NULL;
1407 up_write(&lli->lli_lsm_sem);
1409 GOTO(err_exit, err);
1412 ptlrpc_req_finished(request);
1419 GOTO(err_exit, err);
1421 ll_update_times(request, dir);
1423 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1425 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1427 GOTO(err_exit, err);
1429 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1431 /* must be done before d_instantiate, because it calls
1432 * security_d_instantiate, which means a getxattr if security
1433 * context is not set yet */
1434 err = security_inode_notifysecctx(inode,
1435 op_data->op_file_secctx,
1436 op_data->op_file_secctx_size);
1437 inode_unlock(inode);
1439 GOTO(err_exit, err);
1442 d_instantiate(dchild, inode);
1444 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1445 err = ll_inode_init_security(dchild, inode, dir);
1447 GOTO(err_exit, err);
1452 if (request != NULL)
1453 ptlrpc_req_finished(request);
1455 if (!IS_ERR_OR_NULL(op_data))
1456 ll_finish_md_op_data(op_data);
1461 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1464 struct qstr *name = &dchild->d_name;
1468 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1469 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1472 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1473 mode &= ~current_umask();
1475 switch (mode & S_IFMT) {
1477 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1483 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1494 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1499 #ifdef HAVE_IOP_ATOMIC_OPEN
1501 * Plain create. Intent create is handled in atomic_open.
1503 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1504 umode_t mode, bool want_excl)
1508 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1510 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1511 "flags=%u, excl=%d\n", dentry->d_name.len,
1512 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1513 dir, mode, want_excl);
1515 /* Using mknod(2) to create a regular file is designed to not recognize
1516 * volatile file name, so we use ll_mknod() here. */
1517 rc = ll_mknod(dir, dentry, mode, 0);
1519 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1521 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1522 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1526 #else /* !HAVE_IOP_ATOMIC_OPEN */
1527 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1528 ll_umode_t mode, struct nameidata *nd)
1530 struct ll_dentry_data *lld = ll_d2d(dentry);
1531 struct lookup_intent *it = NULL;
1534 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1540 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1541 * so that volatile file name is recoginized.
1542 * Mknod(2), however, is designed to not recognize volatile
1543 * file name to avoid inode leak under orphan directory until
1545 return ll_new_node(dir, dentry, NULL, mode, 0,
1551 /* Was there an error? Propagate it! */
1552 if (it->it_status) {
1557 rc = ll_create_it(dir, dentry, it, NULL, 0);
1558 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1561 nd->intent.open.file->private_data = it;
1562 filp = lookup_instantiate_filp(nd, dentry, NULL);
1568 ll_intent_release(it);
1569 OBD_FREE(it, sizeof(*it));
1572 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1576 #endif /* HAVE_IOP_ATOMIC_OPEN */
1578 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1579 const char *oldpath)
1581 struct qstr *name = &dchild->d_name;
1585 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1586 name->len, name->name, PFID(ll_inode2fid(dir)),
1587 dir, 3000, oldpath);
1589 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1590 LUSTRE_OPC_SYMLINK);
1593 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1598 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1599 struct dentry *new_dentry)
1601 struct inode *src = old_dentry->d_inode;
1602 struct qstr *name = &new_dentry->d_name;
1603 struct ll_sb_info *sbi = ll_i2sbi(dir);
1604 struct ptlrpc_request *request = NULL;
1605 struct md_op_data *op_data;
1609 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1610 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1611 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1613 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1614 0, LUSTRE_OPC_ANY, NULL);
1615 if (IS_ERR(op_data))
1616 RETURN(PTR_ERR(op_data));
1618 err = md_link(sbi->ll_md_exp, op_data, &request);
1619 ll_finish_md_op_data(op_data);
1623 ll_update_times(request, dir);
1624 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1627 ptlrpc_req_finished(request);
1631 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1633 struct qstr *name = &dchild->d_name;
1637 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1638 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1640 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1641 mode &= ~current_umask();
1643 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1645 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1647 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1652 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1654 struct qstr *name = &dchild->d_name;
1655 struct ptlrpc_request *request = NULL;
1656 struct md_op_data *op_data;
1660 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1661 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1663 if (unlikely(d_mountpoint(dchild)))
1666 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1667 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1668 if (IS_ERR(op_data))
1669 RETURN(PTR_ERR(op_data));
1671 if (dchild->d_inode != NULL)
1672 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1674 op_data->op_fid2 = op_data->op_fid3;
1675 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1676 ll_finish_md_op_data(op_data);
1678 ll_update_times(request, dir);
1679 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1682 ptlrpc_req_finished(request);
1689 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1691 struct ptlrpc_request *request = NULL;
1692 struct md_op_data *op_data;
1696 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1697 namelen, name, PFID(ll_inode2fid(dir)), dir);
1699 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1700 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1701 if (IS_ERR(op_data))
1702 RETURN(PTR_ERR(op_data));
1703 op_data->op_cli_flags |= CLI_RM_ENTRY;
1704 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1705 ll_finish_md_op_data(op_data);
1707 ll_update_times(request, dir);
1708 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1711 ptlrpc_req_finished(request);
1715 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1717 struct qstr *name = &dchild->d_name;
1718 struct ptlrpc_request *request = NULL;
1719 struct md_op_data *op_data;
1720 struct mdt_body *body;
1723 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1724 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1727 * XXX: unlink bind mountpoint maybe call to here,
1728 * just check it as vfs_unlink does.
1730 if (unlikely(d_mountpoint(dchild)))
1733 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1734 LUSTRE_OPC_ANY, NULL);
1735 if (IS_ERR(op_data))
1736 RETURN(PTR_ERR(op_data));
1738 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1740 op_data->op_fid2 = op_data->op_fid3;
1741 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1742 ll_finish_md_op_data(op_data);
1747 * The server puts attributes in on the last unlink, use them to update
1748 * the link count so the inode can be freed immediately.
1750 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1751 if (body->mbo_valid & OBD_MD_FLNLINK)
1752 set_nlink(dchild->d_inode, body->mbo_nlink);
1754 ll_update_times(request, dir);
1755 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1758 ptlrpc_req_finished(request);
1762 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1763 struct inode *tgt, struct dentry *tgt_dchild
1764 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1765 , unsigned int flags
1769 struct qstr *src_name = &src_dchild->d_name;
1770 struct qstr *tgt_name = &tgt_dchild->d_name;
1771 struct ptlrpc_request *request = NULL;
1772 struct ll_sb_info *sbi = ll_i2sbi(src);
1773 struct md_op_data *op_data;
1777 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1782 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1783 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1784 src_name->len, src_name->name,
1785 PFID(ll_inode2fid(src)), src, tgt_name->len,
1786 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1788 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1791 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1792 LUSTRE_OPC_ANY, NULL);
1793 if (IS_ERR(op_data))
1794 RETURN(PTR_ERR(op_data));
1796 if (src_dchild->d_inode != NULL)
1797 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1799 if (tgt_dchild->d_inode != NULL)
1800 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1802 err = md_rename(sbi->ll_md_exp, op_data,
1803 src_name->name, src_name->len,
1804 tgt_name->name, tgt_name->len, &request);
1805 ll_finish_md_op_data(op_data);
1807 ll_update_times(request, src);
1808 ll_update_times(request, tgt);
1809 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1812 ptlrpc_req_finished(request);
1815 d_move(src_dchild, tgt_dchild);
1820 const struct inode_operations ll_dir_inode_operations = {
1822 #ifdef HAVE_IOP_ATOMIC_OPEN
1823 .atomic_open = ll_atomic_open,
1825 .lookup = ll_lookup_nd,
1826 .create = ll_create_nd,
1827 /* We need all these non-raw things for NFSD, to not patch it. */
1828 .unlink = ll_unlink,
1831 .symlink = ll_symlink,
1833 .rename = ll_rename,
1834 .setattr = ll_setattr,
1835 .getattr = ll_getattr,
1836 .permission = ll_inode_permission,
1837 #ifdef HAVE_IOP_XATTR
1838 .setxattr = ll_setxattr,
1839 .getxattr = ll_getxattr,
1840 .removexattr = ll_removexattr,
1842 .listxattr = ll_listxattr,
1843 #ifdef HAVE_IOP_GET_ACL
1844 .get_acl = ll_get_acl,
1846 #ifdef HAVE_IOP_SET_ACL
1847 .set_acl = ll_set_acl,
1851 const struct inode_operations ll_special_inode_operations = {
1852 .setattr = ll_setattr,
1853 .getattr = ll_getattr,
1854 .permission = ll_inode_permission,
1855 #ifdef HAVE_IOP_XATTR
1856 .setxattr = ll_setxattr,
1857 .getxattr = ll_getxattr,
1858 .removexattr = ll_removexattr,
1860 .listxattr = ll_listxattr,
1861 #ifdef HAVE_IOP_GET_ACL
1862 .get_acl = ll_get_acl,
1864 #ifdef HAVE_IOP_SET_ACL
1865 .set_acl = ll_set_acl,