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 /* That means the inode is evicted most likely and may cause
240 * the skipping of lock cleanups below, so print the message
243 if (lock->l_resource->lr_lvb_inode)
245 "can't take inode for the lock (%sevicted)\n",
246 lock->l_resource->lr_lvb_inode->i_state &
247 I_FREEING ? "" : "not ");
251 if (!fid_res_name_eq(ll_inode2fid(inode),
252 &lock->l_resource->lr_name)) {
253 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
254 PFID(ll_inode2fid(inode)), inode);
258 if (bits & MDS_INODELOCK_XATTR) {
259 if (S_ISDIR(inode->i_mode))
260 ll_i2info(inode)->lli_def_stripe_offset = -1;
261 ll_xattr_cache_destroy(inode);
262 bits &= ~MDS_INODELOCK_XATTR;
265 /* For OPEN locks we differentiate between lock modes
266 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
267 if (bits & MDS_INODELOCK_OPEN)
268 ll_have_md_lock(inode, &bits, lock->l_req_mode);
270 if (bits & MDS_INODELOCK_OPEN) {
273 switch (lock->l_req_mode) {
284 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
288 ll_md_real_close(inode, fmode);
290 bits &= ~MDS_INODELOCK_OPEN;
293 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
294 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
296 ll_have_md_lock(inode, &bits, LCK_MINMODE);
298 if (bits & MDS_INODELOCK_DOM) {
299 rc = ll_dom_lock_cancel(inode, lock);
301 CDEBUG(D_INODE, "cannot flush DoM data "
303 PFID(ll_inode2fid(inode)), rc);
304 lock_res_and_lock(lock);
305 ldlm_set_kms_ignore(lock);
306 unlock_res_and_lock(lock);
309 if (bits & MDS_INODELOCK_LAYOUT) {
310 struct cl_object_conf conf = {
311 .coc_opc = OBJECT_CONF_INVALIDATE,
315 rc = ll_layout_conf(inode, &conf);
317 CDEBUG(D_INODE, "cannot invalidate layout of "
319 PFID(ll_inode2fid(inode)), rc);
322 if (bits & MDS_INODELOCK_UPDATE) {
323 struct ll_inode_info *lli = ll_i2info(inode);
325 lli->lli_update_atime = 1;
328 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
329 struct ll_inode_info *lli = ll_i2info(inode);
331 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
332 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
333 lli, PFID(&lli->lli_pfid));
334 truncate_inode_pages(inode->i_mapping, 0);
336 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
337 struct inode *master_inode = NULL;
340 /* This is slave inode, since all of the child dentry
341 * is connected on the master inode, so we have to
342 * invalidate the negative children on master inode */
343 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
344 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
346 hash = cl_fid_build_ino(&lli->lli_pfid,
347 ll_need_32bit_api(ll_i2sbi(inode)));
349 /* Do not lookup the inode with ilookup5, otherwise
350 * it will cause dead lock,
351 * 1. Client1 send chmod req to the MDT0, then on MDT0,
352 * it enqueues master and all of its slaves lock,
353 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
354 * master and stripe0 lock, it will send the enqueue
355 * req (for stripe1) to MDT1, then MDT1 finds the lock
356 * has been granted to client2. Then MDT1 sends blocking
358 * 2. At the same time, client2 tries to unlink
359 * the striped dir (rm -rf striped_dir), and during
360 * lookup, it will hold the master inode of the striped
361 * directory, whose inode state is NEW, then tries to
362 * revalidate all of its slaves, (ll_prep_inode()->
363 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
364 * And it will be blocked on the server side because
366 * 3. Then the client get the blocking_ast req, cancel
367 * the lock, but being blocked if using ->ilookup5()),
368 * because master inode state is NEW. */
369 master_inode = ilookup5_nowait(inode->i_sb, hash,
370 ll_test_inode_by_fid,
371 (void *)&lli->lli_pfid);
373 ll_invalidate_negative_children(master_inode);
377 ll_invalidate_negative_children(inode);
381 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
382 inode->i_sb->s_root != NULL &&
383 inode != inode->i_sb->s_root->d_inode)
384 ll_invalidate_aliases(inode);
390 /* Check if the given lock may be downgraded instead of canceling and
391 * that convert is really needed. */
392 int ll_md_need_convert(struct ldlm_lock *lock)
394 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
396 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
397 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
398 enum ldlm_mode mode = LCK_MINMODE;
400 if (!lock->l_conn_export ||
401 !exp_connect_lock_convert(lock->l_conn_export))
404 if (!wanted || !bits || ldlm_is_cancel(lock))
407 /* do not convert locks other than DOM for now */
408 if (!((bits | wanted) & MDS_INODELOCK_DOM))
411 /* We may have already remaining bits in some other lock so
412 * lock convert will leave us just extra lock for the same bit.
413 * Check if client has other lock with the same bits and the same
414 * or lower mode and don't convert if any.
416 switch (lock->l_req_mode) {
428 /* do not convert other modes */
432 /* is lock is too old to be converted? */
433 lock_res_and_lock(lock);
434 if (ktime_after(ktime_get(),
435 ktime_add(lock->l_last_used,
436 ktime_set(ns->ns_dirty_age_limit, 0)))) {
437 unlock_res_and_lock(lock);
440 unlock_res_and_lock(lock);
442 inode = ll_inode_from_resource_lock(lock);
443 ll_have_md_lock(inode, &bits, mode);
448 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
449 void *data, int flag)
451 struct lustre_handle lockh;
452 __u64 bits = lock->l_policy_data.l_inodebits.bits;
458 case LDLM_CB_BLOCKING:
460 __u64 cancel_flags = LCF_ASYNC;
462 if (ll_md_need_convert(lock)) {
463 cancel_flags |= LCF_CONVERT;
464 /* For lock convert some cancel actions may require
465 * this lock with non-dropped canceled bits, e.g. page
466 * flush for DOM lock. So call ll_lock_cancel_bits()
467 * here while canceled bits are still set.
469 bits = lock->l_policy_data.l_inodebits.cancel_bits;
470 if (bits & MDS_INODELOCK_DOM)
471 ll_lock_cancel_bits(lock, MDS_INODELOCK_DOM);
473 ldlm_lock2handle(lock, &lockh);
474 rc = ldlm_cli_cancel(&lockh, cancel_flags);
476 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
481 case LDLM_CB_CANCELING:
482 /* Nothing to do for non-granted locks */
483 if (!ldlm_is_granted(lock))
486 if (ldlm_is_converting(lock)) {
487 /* this is called on already converted lock, so
488 * ibits has remained bits only and cancel_bits
489 * are bits that were dropped.
490 * Note that DOM lock is handled prior lock convert
491 * and is excluded here.
493 bits = lock->l_policy_data.l_inodebits.cancel_bits &
496 LASSERT(ldlm_is_canceling(lock));
498 ll_lock_cancel_bits(lock, bits);
507 __u32 ll_i2suppgid(struct inode *i)
509 if (in_group_p(i->i_gid))
510 return (__u32)from_kgid(&init_user_ns, i->i_gid);
512 return (__u32) __kgid_val(INVALID_GID);
515 /* Pack the required supplementary groups into the supplied groups array.
516 * If we don't need to use the groups from the target inode(s) then we
517 * instead pack one or more groups from the user's supplementary group
518 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
519 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
522 LASSERT(suppgids != NULL);
524 suppgids[0] = ll_i2suppgid(i1);
527 suppgids[1] = ll_i2suppgid(i2);
533 * try to reuse three types of dentry:
534 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
535 * by concurrent .revalidate).
536 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
537 * be cleared by others calling d_lustre_revalidate).
538 * 3. DISCONNECTED alias.
540 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
542 struct dentry *alias, *discon_alias, *invalid_alias;
543 DECLARE_LL_D_HLIST_NODE_PTR(p);
545 if (ll_d_hlist_empty(&inode->i_dentry))
548 discon_alias = invalid_alias = NULL;
550 ll_lock_dcache(inode);
551 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
552 LASSERT(alias != dentry);
554 spin_lock(&alias->d_lock);
555 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
556 S_ISDIR(inode->i_mode))
557 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
558 discon_alias = alias;
559 else if (alias->d_parent == dentry->d_parent &&
560 alias->d_name.hash == dentry->d_name.hash &&
561 alias->d_name.len == dentry->d_name.len &&
562 memcmp(alias->d_name.name, dentry->d_name.name,
563 dentry->d_name.len) == 0)
564 invalid_alias = alias;
565 spin_unlock(&alias->d_lock);
570 alias = invalid_alias ?: discon_alias ?: NULL;
572 spin_lock(&alias->d_lock);
574 spin_unlock(&alias->d_lock);
576 ll_unlock_dcache(inode);
582 * Similar to d_splice_alias(), but lustre treats invalid alias
583 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
585 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
591 new = ll_find_alias(inode, de);
601 "Reuse dentry %p inode %p refc %d flags %#x\n",
602 new, new->d_inode, ll_d_count(new), new->d_flags);
610 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
611 de, de->d_inode, ll_d_count(de), de->d_flags);
615 static int ll_lookup_it_finish(struct ptlrpc_request *request,
616 struct lookup_intent *it,
617 struct inode *parent, struct dentry **de,
618 void *secctx, __u32 secctxlen)
620 struct inode *inode = NULL;
623 struct dentry *alias;
626 /* NB 1 request reference will be taken away by ll_intent_lock()
628 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
630 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
631 struct req_capsule *pill = &request->rq_pill;
632 struct mdt_body *body = req_capsule_server_get(pill,
635 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
639 if (it->it_op & IT_OPEN)
640 ll_dom_finish_open(inode, request, it);
642 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
644 /* We used to query real size from OSTs here, but actually
645 * this is not needed. For stat() calls size would be updated
646 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
648 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
649 * Everybody else who needs correct file size would call
650 * ll_glimpse_size or some equivalent themselves anyway.
654 /* If security context was returned by MDT, put it in
655 * inode now to save an extra getxattr from security hooks,
656 * and avoid deadlock.
658 if (body->mbo_valid & OBD_MD_SECCTX) {
659 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
660 secctxlen = req_capsule_get_size(pill,
665 CDEBUG(D_SEC, "server returned security context"
667 PFID(ll_inode2fid(inode)));
670 if (secctx != NULL && secctxlen != 0) {
672 rc = security_inode_notifysecctx(inode, secctx,
676 CWARN("cannot set security context for "
678 PFID(ll_inode2fid(inode)), rc);
682 /* Only hash *de if it is unhashed (new dentry).
683 * Atoimc_open may passin hashed dentries for open.
685 alias = ll_splice_alias(inode, *de);
687 GOTO(out, rc = PTR_ERR(alias));
691 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
692 /* we have lookup look - unhide dentry */
693 if (bits & MDS_INODELOCK_LOOKUP)
694 d_lustre_revalidate(*de);
695 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
697 * If file was created on the server, the dentry is revalidated
698 * in ll_create_it if the lock allows for it.
700 /* Check that parent has UPDATE lock. */
701 struct lookup_intent parent_it = {
703 .it_lock_handle = 0 };
704 struct lu_fid fid = ll_i2info(parent)->lli_fid;
706 /* If it is striped directory, get the real stripe parent */
707 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
708 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
709 ll_i2info(parent)->lli_lsm_md,
711 (*de)->d_name.len, &fid);
716 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
718 d_lustre_revalidate(*de);
719 ll_intent_release(&parent_it);
726 if (rc != 0 && it->it_op & IT_OPEN)
727 ll_open_cleanup((*de)->d_sb, request);
732 struct pcc_create_attach {
733 struct pcc_dataset *pca_dataset;
734 struct dentry *pca_dentry;
737 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
738 struct lookup_intent *it,
739 void **secctx, __u32 *secctxlen,
740 struct pcc_create_attach *pca)
742 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
743 struct dentry *save = dentry, *retval;
744 struct ptlrpc_request *req = NULL;
745 struct md_op_data *op_data = NULL;
746 struct lov_user_md *lum = NULL;
749 char secctx_name[XATTR_NAME_MAX + 1];
753 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
754 RETURN(ERR_PTR(-ENAMETOOLONG));
756 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
757 dentry->d_name.len, dentry->d_name.name,
758 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
760 if (d_mountpoint(dentry))
761 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
763 if (it == NULL || it->it_op == IT_GETXATTR)
766 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
767 rc = ll_statahead(parent, &dentry, 0);
769 RETURN(dentry == save ? NULL : dentry);
772 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
773 dentry->d_sb->s_flags & MS_RDONLY)
774 RETURN(ERR_PTR(-EROFS));
776 if (it->it_op & IT_CREAT)
777 opc = LUSTRE_OPC_CREATE;
779 opc = LUSTRE_OPC_ANY;
781 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
782 dentry->d_name.len, 0, opc, NULL);
784 GOTO(out, retval = ERR_CAST(op_data));
786 /* enforce umask if acl disabled or MDS doesn't support umask */
787 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
788 it->it_create_mode &= ~current_umask();
790 if (it->it_op & IT_CREAT &&
791 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
792 rc = ll_dentry_init_security(dentry, it->it_create_mode,
794 &op_data->op_file_secctx_name,
795 &op_data->op_file_secctx,
796 &op_data->op_file_secctx_size);
798 GOTO(out, retval = ERR_PTR(rc));
800 *secctx = op_data->op_file_secctx;
801 if (secctxlen != NULL)
802 *secctxlen = op_data->op_file_secctx_size;
806 if (secctxlen != NULL)
810 /* ask for security context upon intent */
811 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
812 /* get name of security xattr to request to server */
813 rc = ll_listsecurity(parent, secctx_name,
814 sizeof(secctx_name));
816 CDEBUG(D_SEC, "cannot get security xattr name for "
818 PFID(ll_inode2fid(parent)), rc);
820 op_data->op_file_secctx_name = secctx_name;
821 op_data->op_file_secctx_name_size = rc;
822 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
823 rc, secctx_name, PFID(ll_inode2fid(parent)));
827 if (pca && pca->pca_dataset) {
828 struct pcc_dataset *dataset = pca->pca_dataset;
832 GOTO(out, retval = ERR_PTR(-ENOMEM));
834 lum->lmm_magic = LOV_USER_MAGIC_V1;
835 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
836 lum->lmm_stripe_size = 0;
837 lum->lmm_stripe_count = 0;
838 lum->lmm_stripe_offset = 0;
840 op_data->op_data = lum;
841 op_data->op_data_size = sizeof(*lum);
842 op_data->op_archive_id = dataset->pccd_id;
844 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
847 GOTO(out, retval = ERR_PTR(rc));
849 rc = pcc_inode_create(dataset, &op_data->op_fid2,
852 GOTO(out, retval = ERR_PTR(rc));
854 it->it_flags |= MDS_OPEN_PCC;
857 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
858 &ll_md_blocking_ast, 0);
859 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
860 * client does not know which suppgid should be sent to the MDS, or
861 * some other(s) changed the target file's GID after this RPC sent
862 * to the MDS with the suppgid as the original GID, then we should
863 * try again with right suppgid. */
864 if (rc == -EACCES && it->it_op & IT_OPEN &&
865 it_disposition(it, DISP_OPEN_DENY)) {
866 struct mdt_body *body;
868 LASSERT(req != NULL);
870 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
871 if (op_data->op_suppgids[0] == body->mbo_gid ||
872 op_data->op_suppgids[1] == body->mbo_gid ||
873 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
874 GOTO(out, retval = ERR_PTR(-EACCES));
876 fid_zero(&op_data->op_fid2);
877 op_data->op_suppgids[1] = body->mbo_gid;
878 ptlrpc_req_finished(req);
880 ll_intent_release(it);
881 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
882 &ll_md_blocking_ast, 0);
886 GOTO(out, retval = ERR_PTR(rc));
888 /* dir layout may change */
889 ll_unlock_md_op_lsm(op_data);
890 rc = ll_lookup_it_finish(req, it, parent, &dentry,
891 secctx != NULL ? *secctx : NULL,
892 secctxlen != NULL ? *secctxlen : 0);
894 ll_intent_release(it);
895 GOTO(out, retval = ERR_PTR(rc));
898 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
899 !S_ISREG(dentry->d_inode->i_mode) &&
900 !S_ISDIR(dentry->d_inode->i_mode)) {
901 ll_release_openhandle(dentry, it);
903 ll_lookup_finish_locks(it, dentry);
905 GOTO(out, retval = (dentry == save) ? NULL : dentry);
908 if (op_data != NULL && !IS_ERR(op_data)) {
909 if (secctx != NULL && secctxlen != NULL) {
910 /* caller needs sec ctx info, so reset it in op_data to
911 * prevent it from being freed */
912 op_data->op_file_secctx = NULL;
913 op_data->op_file_secctx_size = 0;
915 ll_finish_md_op_data(op_data);
921 ptlrpc_req_finished(req);
925 #ifdef HAVE_IOP_ATOMIC_OPEN
926 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
929 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
932 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
933 dentry->d_name.len, dentry->d_name.name,
934 PFID(ll_inode2fid(parent)), parent, flags);
937 * Optimize away (CREATE && !OPEN). Let .create handle the race.
938 * but only if we have write permissions there, otherwise we need
939 * to proceed with lookup. LU-4185
941 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
942 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
945 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
949 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL);
952 ll_intent_release(itp);
958 * For cached negative dentry and new dentry, handle lookup/create/open
961 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
962 struct file *file, unsigned open_flags,
963 umode_t mode, int *opened)
965 struct lookup_intent *it;
967 long long lookup_flags = LOOKUP_OPEN;
970 struct ll_sb_info *sbi;
971 struct pcc_create_attach pca = {NULL, NULL};
972 struct pcc_dataset *dataset = NULL;
976 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
977 "open_flags %x, mode %x opened %d\n",
978 dentry->d_name.len, dentry->d_name.name,
979 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
981 /* Only negative dentries enter here */
982 LASSERT(dentry->d_inode == NULL);
984 if (!d_unhashed(dentry)) {
985 /* A valid negative dentry that just passed revalidation,
986 * there's little point to try and open it server-side,
987 * even though there's a minuscule chance it might succeed.
988 * Either way it's a valid race to just return -ENOENT here.
990 if (!(open_flags & O_CREAT))
993 /* Otherwise we just unhash it to be rehashed afresh via
994 * lookup if necessary
999 OBD_ALLOC(it, sizeof(*it));
1003 it->it_op = IT_OPEN;
1004 if (open_flags & O_CREAT) {
1005 it->it_op |= IT_CREAT;
1006 lookup_flags |= LOOKUP_CREATE;
1007 sbi = ll_i2sbi(dir);
1008 /* Volatile file is used for HSM restore, so do not use PCC */
1009 if (!filename_is_volatile(dentry->d_name.name,
1010 dentry->d_name.len, NULL)) {
1011 dataset = pcc_dataset_get(&sbi->ll_pcc_super,
1012 ll_i2info(dir)->lli_projid,
1014 pca.pca_dataset = dataset;
1017 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1018 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1019 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1021 /* Dentry added to dcache tree in ll_lookup_it */
1022 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca);
1025 else if (de != NULL)
1028 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1031 if (it_disposition(it, DISP_OPEN_CREATE)) {
1032 /* Dentry instantiated in ll_create_it. */
1033 rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
1034 security_release_secctx(secctx, secctxlen);
1036 /* We dget in ll_splice_alias. */
1041 if (dataset != NULL && dentry->d_inode) {
1042 rc = pcc_inode_create_fini(dataset,
1048 GOTO(out_release, rc);
1052 *opened |= FILE_CREATED;
1055 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1057 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1058 /* We cannot call open here as it might
1059 * deadlock. This case is unreachable in
1060 * practice because of OBD_CONNECT_NODEVOH. */
1061 rc = finish_no_open(file, de);
1063 file->private_data = it;
1064 rc = finish_open(file, dentry, NULL, opened);
1065 /* We dget in ll_splice_alias. finish_open takes
1066 * care of dget for fd open.
1072 rc = finish_no_open(file, de);
1077 if (dataset != NULL)
1078 pcc_dataset_put(dataset);
1079 ll_intent_release(it);
1080 OBD_FREE(it, sizeof(*it));
1085 #else /* !HAVE_IOP_ATOMIC_OPEN */
1086 static struct lookup_intent *
1087 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1089 struct lookup_intent *it;
1093 return ERR_PTR(-ENOMEM);
1095 if (lookup_flags & LOOKUP_OPEN) {
1096 it->it_op = IT_OPEN;
1097 /* Avoid file creation for ro bind mount point(is_readonly) */
1098 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1099 it->it_op |= IT_CREAT;
1100 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1101 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1102 ~(is_readonly ? O_CREAT : 0));
1103 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1105 it->it_op = IT_GETATTR;
1111 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1112 struct nameidata *nd)
1117 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1118 struct lookup_intent *it;
1120 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1121 it = ll_d2d(dentry)->lld_it;
1122 ll_d2d(dentry)->lld_it = NULL;
1125 * Optimize away (CREATE && !OPEN). Let .create handle
1126 * the race. But only if we have write permissions
1127 * there, otherwise we need to proceed with lookup.
1130 if ((nd->flags & LOOKUP_CREATE) &&
1131 !(nd->flags & LOOKUP_OPEN) &&
1132 (inode_permission(parent,
1133 MAY_WRITE | MAY_EXEC) == 0))
1136 it = ll_convert_intent(&nd->intent.open, nd->flags,
1137 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1138 (nd->path.mnt->mnt_sb->s_flags & MS_RDONLY));
1140 RETURN((struct dentry *)it);
1143 de = ll_lookup_it(parent, dentry, it, NULL, NULL, NULL);
1146 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1147 if (dentry->d_inode &&
1148 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1149 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1150 /* We cannot call open here as it might
1151 * deadlock. This case is unreachable in
1152 * practice because of
1153 * OBD_CONNECT_NODEVOH. */
1157 nd->intent.open.file->private_data = it;
1158 filp = lookup_instantiate_filp(nd,
1164 de = (struct dentry *)filp;
1167 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1168 /* XXX This can only reliably work on assumption
1169 * that there are NO hashed negative dentries.*/
1170 ll_d2d(dentry)->lld_it = it;
1171 it = NULL; /* Will be freed in ll_create_nd */
1172 /* We absolutely depend on ll_create_nd to be
1173 * called to not leak this intent and possible
1174 * data attached to it */
1179 ll_intent_release(it);
1180 OBD_FREE(it, sizeof(*it));
1183 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL, NULL);
1188 #endif /* HAVE_IOP_ATOMIC_OPEN */
1190 /* We depend on "mode" being set with the proper file type/umask by now */
1191 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1193 struct inode *inode = NULL;
1194 struct ptlrpc_request *request = NULL;
1195 struct ll_sb_info *sbi = ll_i2sbi(dir);
1199 LASSERT(it && it->it_disposition);
1201 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1202 request = it->it_request;
1203 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1204 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1206 GOTO(out, inode = ERR_PTR(rc));
1208 /* Pause to allow for a race with concurrent access by fid */
1209 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1211 /* We asked for a lock on the directory, but were granted a
1212 * lock on the inode. Since we finally have an inode pointer,
1213 * stuff it in the lock. */
1214 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1215 PFID(ll_inode2fid(inode)), inode);
1216 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1219 ptlrpc_req_finished(request);
1224 * By the time this is called, we already have created the directory cache
1225 * entry for the new file, but it is so far negative - it has no inode.
1227 * We defer creating the OBD object(s) until open, to keep the intent and
1228 * non-intent code paths similar, and also because we do not have the MDS
1229 * inode number before calling ll_create_node() (which is needed for LOV),
1230 * so we would need to do yet another RPC to the MDS to store the LOV EA
1231 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1232 * lmm_size in datalen (the MDS still has code which will handle that).
1234 * If the create succeeds, we fill in the inode information
1235 * with d_instantiate().
1237 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1238 struct lookup_intent *it,
1239 void *secctx, __u32 secctxlen)
1241 struct inode *inode;
1246 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1247 dentry->d_name.len, dentry->d_name.name,
1248 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1250 rc = it_open_error(DISP_OPEN_CREATE, it);
1254 inode = ll_create_node(dir, it);
1256 RETURN(PTR_ERR(inode));
1258 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1261 /* must be done before d_instantiate, because it calls
1262 * security_d_instantiate, which means a getxattr if security
1263 * context is not set yet */
1264 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1265 inode_unlock(inode);
1270 d_instantiate(dentry, inode);
1272 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1273 rc = ll_inode_init_security(dentry, inode, dir);
1278 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1279 if (bits & MDS_INODELOCK_LOOKUP)
1280 d_lustre_revalidate(dentry);
1285 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1287 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1291 if (body->mbo_valid & OBD_MD_FLMTIME &&
1292 body->mbo_mtime > inode->i_mtime.tv_sec) {
1294 "setting fid " DFID " mtime from %lld to %llu\n",
1295 PFID(ll_inode2fid(inode)),
1296 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1297 inode->i_mtime.tv_sec = body->mbo_mtime;
1300 if (body->mbo_valid & OBD_MD_FLCTIME &&
1301 body->mbo_ctime > inode->i_ctime.tv_sec)
1302 inode->i_ctime.tv_sec = body->mbo_ctime;
1305 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1306 const char *tgt, umode_t mode, int rdev, __u32 opc)
1308 struct qstr *name = &dchild->d_name;
1309 struct ptlrpc_request *request = NULL;
1310 struct md_op_data *op_data;
1311 struct inode *inode = NULL;
1312 struct ll_sb_info *sbi = ll_i2sbi(dir);
1317 if (unlikely(tgt != NULL))
1318 tgt_len = strlen(tgt) + 1;
1321 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1322 name->len, 0, opc, NULL);
1323 if (IS_ERR(op_data))
1324 GOTO(err_exit, err = PTR_ERR(op_data));
1326 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1327 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1328 &op_data->op_file_secctx_name,
1329 &op_data->op_file_secctx,
1330 &op_data->op_file_secctx_size);
1332 GOTO(err_exit, err);
1335 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1336 from_kuid(&init_user_ns, current_fsuid()),
1337 from_kgid(&init_user_ns, current_fsgid()),
1338 cfs_curproc_cap_pack(), rdev, &request);
1339 if (err < 0 && err != -EREMOTE)
1340 GOTO(err_exit, err);
1342 /* If the client doesn't know where to create a subdirectory (or
1343 * in case of a race that sends the RPC to the wrong MDS), the
1344 * MDS will return -EREMOTE and the client will fetch the layout
1345 * of the directory, then create the directory on the right MDT. */
1346 if (unlikely(err == -EREMOTE)) {
1347 struct ll_inode_info *lli = ll_i2info(dir);
1348 struct lmv_user_md *lum;
1352 ptlrpc_req_finished(request);
1355 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1356 OBD_MD_DEFAULT_MEA);
1358 /* Update stripe_offset and retry */
1359 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1360 } else if (err2 == -ENODATA &&
1361 lli->lli_def_stripe_offset != -1) {
1362 /* If there are no default stripe EA on the MDT, but the
1363 * client has default stripe, then it probably means
1364 * default stripe EA has just been deleted. */
1365 lli->lli_def_stripe_offset = -1;
1367 GOTO(err_exit, err);
1370 ptlrpc_req_finished(request);
1372 ll_finish_md_op_data(op_data);
1376 ll_update_times(request, dir);
1378 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1380 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1382 GOTO(err_exit, err);
1384 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1386 /* must be done before d_instantiate, because it calls
1387 * security_d_instantiate, which means a getxattr if security
1388 * context is not set yet */
1389 err = security_inode_notifysecctx(inode,
1390 op_data->op_file_secctx,
1391 op_data->op_file_secctx_size);
1392 inode_unlock(inode);
1394 GOTO(err_exit, err);
1397 d_instantiate(dchild, inode);
1399 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1400 err = ll_inode_init_security(dchild, inode, dir);
1402 GOTO(err_exit, err);
1407 if (request != NULL)
1408 ptlrpc_req_finished(request);
1410 if (!IS_ERR_OR_NULL(op_data))
1411 ll_finish_md_op_data(op_data);
1416 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1419 struct qstr *name = &dchild->d_name;
1423 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1424 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1427 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1428 mode &= ~current_umask();
1430 switch (mode & S_IFMT) {
1432 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1438 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1449 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1454 #ifdef HAVE_IOP_ATOMIC_OPEN
1456 * Plain create. Intent create is handled in atomic_open.
1458 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1459 umode_t mode, bool want_excl)
1463 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1465 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1466 "flags=%u, excl=%d\n", dentry->d_name.len,
1467 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1468 dir, mode, want_excl);
1470 /* Using mknod(2) to create a regular file is designed to not recognize
1471 * volatile file name, so we use ll_mknod() here. */
1472 rc = ll_mknod(dir, dentry, mode, 0);
1474 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1476 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1477 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1481 #else /* !HAVE_IOP_ATOMIC_OPEN */
1482 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1483 ll_umode_t mode, struct nameidata *nd)
1485 struct ll_dentry_data *lld = ll_d2d(dentry);
1486 struct lookup_intent *it = NULL;
1489 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1495 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1496 * so that volatile file name is recoginized.
1497 * Mknod(2), however, is designed to not recognize volatile
1498 * file name to avoid inode leak under orphan directory until
1500 return ll_new_node(dir, dentry, NULL, mode, 0,
1506 /* Was there an error? Propagate it! */
1507 if (it->it_status) {
1512 rc = ll_create_it(dir, dentry, it, NULL, 0);
1513 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1516 nd->intent.open.file->private_data = it;
1517 filp = lookup_instantiate_filp(nd, dentry, NULL);
1523 ll_intent_release(it);
1524 OBD_FREE(it, sizeof(*it));
1527 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1531 #endif /* HAVE_IOP_ATOMIC_OPEN */
1533 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1534 const char *oldpath)
1536 struct qstr *name = &dchild->d_name;
1540 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1541 name->len, name->name, PFID(ll_inode2fid(dir)),
1542 dir, 3000, oldpath);
1544 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1545 LUSTRE_OPC_SYMLINK);
1548 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1553 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1554 struct dentry *new_dentry)
1556 struct inode *src = old_dentry->d_inode;
1557 struct qstr *name = &new_dentry->d_name;
1558 struct ll_sb_info *sbi = ll_i2sbi(dir);
1559 struct ptlrpc_request *request = NULL;
1560 struct md_op_data *op_data;
1564 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1565 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1566 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1568 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1569 0, LUSTRE_OPC_ANY, NULL);
1570 if (IS_ERR(op_data))
1571 RETURN(PTR_ERR(op_data));
1573 err = md_link(sbi->ll_md_exp, op_data, &request);
1574 ll_finish_md_op_data(op_data);
1578 ll_update_times(request, dir);
1579 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1582 ptlrpc_req_finished(request);
1586 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1588 struct qstr *name = &dchild->d_name;
1592 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1593 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1595 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1596 mode &= ~current_umask();
1598 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1600 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1602 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1607 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1609 struct qstr *name = &dchild->d_name;
1610 struct ptlrpc_request *request = NULL;
1611 struct md_op_data *op_data;
1615 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1616 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1618 if (unlikely(d_mountpoint(dchild)))
1621 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1622 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1623 if (IS_ERR(op_data))
1624 RETURN(PTR_ERR(op_data));
1626 if (dchild->d_inode != NULL)
1627 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1629 op_data->op_fid2 = op_data->op_fid3;
1630 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1631 ll_finish_md_op_data(op_data);
1633 ll_update_times(request, dir);
1634 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1637 ptlrpc_req_finished(request);
1644 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1646 struct ptlrpc_request *request = NULL;
1647 struct md_op_data *op_data;
1651 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1652 namelen, name, PFID(ll_inode2fid(dir)), dir);
1654 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1655 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1656 if (IS_ERR(op_data))
1657 RETURN(PTR_ERR(op_data));
1658 op_data->op_cli_flags |= CLI_RM_ENTRY;
1659 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1660 ll_finish_md_op_data(op_data);
1662 ll_update_times(request, dir);
1663 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1666 ptlrpc_req_finished(request);
1670 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1672 struct qstr *name = &dchild->d_name;
1673 struct ptlrpc_request *request = NULL;
1674 struct md_op_data *op_data;
1675 struct mdt_body *body;
1678 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1679 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1682 * XXX: unlink bind mountpoint maybe call to here,
1683 * just check it as vfs_unlink does.
1685 if (unlikely(d_mountpoint(dchild)))
1688 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1689 LUSTRE_OPC_ANY, NULL);
1690 if (IS_ERR(op_data))
1691 RETURN(PTR_ERR(op_data));
1693 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1695 op_data->op_fid2 = op_data->op_fid3;
1696 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1697 ll_finish_md_op_data(op_data);
1702 * The server puts attributes in on the last unlink, use them to update
1703 * the link count so the inode can be freed immediately.
1705 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1706 if (body->mbo_valid & OBD_MD_FLNLINK)
1707 set_nlink(dchild->d_inode, body->mbo_nlink);
1709 ll_update_times(request, dir);
1710 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1713 ptlrpc_req_finished(request);
1717 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1718 struct inode *tgt, struct dentry *tgt_dchild
1719 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1720 , unsigned int flags
1724 struct qstr *src_name = &src_dchild->d_name;
1725 struct qstr *tgt_name = &tgt_dchild->d_name;
1726 struct ptlrpc_request *request = NULL;
1727 struct ll_sb_info *sbi = ll_i2sbi(src);
1728 struct md_op_data *op_data;
1732 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1737 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1738 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1739 src_name->len, src_name->name,
1740 PFID(ll_inode2fid(src)), src, tgt_name->len,
1741 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1743 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1746 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1747 LUSTRE_OPC_ANY, NULL);
1748 if (IS_ERR(op_data))
1749 RETURN(PTR_ERR(op_data));
1751 if (src_dchild->d_inode != NULL)
1752 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1754 if (tgt_dchild->d_inode != NULL)
1755 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1757 err = md_rename(sbi->ll_md_exp, op_data,
1758 src_name->name, src_name->len,
1759 tgt_name->name, tgt_name->len, &request);
1760 ll_finish_md_op_data(op_data);
1762 ll_update_times(request, src);
1763 ll_update_times(request, tgt);
1764 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1767 ptlrpc_req_finished(request);
1770 d_move(src_dchild, tgt_dchild);
1775 const struct inode_operations ll_dir_inode_operations = {
1777 #ifdef HAVE_IOP_ATOMIC_OPEN
1778 .atomic_open = ll_atomic_open,
1780 .lookup = ll_lookup_nd,
1781 .create = ll_create_nd,
1782 /* We need all these non-raw things for NFSD, to not patch it. */
1783 .unlink = ll_unlink,
1786 .symlink = ll_symlink,
1788 .rename = ll_rename,
1789 .setattr = ll_setattr,
1790 .getattr = ll_getattr,
1791 .permission = ll_inode_permission,
1792 #ifdef HAVE_IOP_XATTR
1793 .setxattr = ll_setxattr,
1794 .getxattr = ll_getxattr,
1795 .removexattr = ll_removexattr,
1797 .listxattr = ll_listxattr,
1798 #ifdef HAVE_IOP_GET_ACL
1799 .get_acl = ll_get_acl,
1801 #ifdef HAVE_IOP_SET_ACL
1802 .set_acl = ll_set_acl,
1806 const struct inode_operations ll_special_inode_operations = {
1807 .setattr = ll_setattr,
1808 .getattr = ll_getattr,
1809 .permission = ll_inode_permission,
1810 #ifdef HAVE_IOP_XATTR
1811 .setxattr = ll_setxattr,
1812 .getxattr = ll_getxattr,
1813 .removexattr = ll_removexattr,
1815 .listxattr = ll_listxattr,
1816 #ifdef HAVE_IOP_GET_ACL
1817 .get_acl = ll_get_acl,
1819 #ifdef HAVE_IOP_SET_ACL
1820 .set_acl = ll_set_acl,