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, 2016, 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 <lustre_ver.h>
51 #include "llite_internal.h"
53 static int ll_create_it(struct inode *dir, struct dentry *dentry,
54 struct lookup_intent *it,
55 void *secctx, __u32 secctxlen);
57 /* called from iget5_locked->find_inode() under inode_lock spinlock */
58 static int ll_test_inode(struct inode *inode, void *opaque)
60 struct ll_inode_info *lli = ll_i2info(inode);
61 struct lustre_md *md = opaque;
63 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
64 CERROR("MDS body missing FID\n");
68 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
74 static int ll_set_inode(struct inode *inode, void *opaque)
76 struct ll_inode_info *lli = ll_i2info(inode);
77 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
79 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
80 CERROR("MDS body missing FID\n");
84 lli->lli_fid = body->mbo_fid1;
85 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
86 CERROR("Can not initialize inode "DFID" without object type: "
88 PFID(&lli->lli_fid), body->mbo_valid);
92 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
93 if (unlikely(inode->i_mode == 0)) {
94 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
105 * Get an inode by inode number(@hash), which is already instantiated by
106 * the intent lookup).
108 struct inode *ll_iget(struct super_block *sb, ino_t hash,
109 struct lustre_md *md)
117 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
119 RETURN(ERR_PTR(-ENOMEM));
121 if (inode->i_state & I_NEW) {
122 rc = ll_read_inode2(inode, md);
123 if (rc == 0 && S_ISREG(inode->i_mode) &&
124 ll_i2info(inode)->lli_clob == NULL)
125 rc = cl_file_inode_init(inode, md);
128 /* Let's clear directory lsm here, otherwise
129 * make_bad_inode() will reset the inode mode
130 * to regular, then ll_clear_inode will not
131 * be able to clear lsm_md */
132 if (S_ISDIR(inode->i_mode))
133 ll_dir_clear_lsm_md(inode);
134 make_bad_inode(inode);
135 unlock_new_inode(inode);
139 unlock_new_inode(inode);
141 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
142 rc = ll_update_inode(inode, md);
143 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
144 PFID(&md->body->mbo_fid1), inode, rc);
146 if (S_ISDIR(inode->i_mode))
147 ll_dir_clear_lsm_md(inode);
156 static void ll_invalidate_negative_children(struct inode *dir)
158 struct dentry *dentry, *tmp_subdir;
159 DECLARE_LL_D_HLIST_NODE_PTR(p);
162 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
163 spin_lock(&dentry->d_lock);
164 if (!list_empty(&dentry->d_subdirs)) {
165 struct dentry *child;
167 list_for_each_entry_safe(child, tmp_subdir,
170 if (child->d_inode == NULL)
171 d_lustre_invalidate(child, 1);
174 spin_unlock(&dentry->d_lock);
176 ll_unlock_dcache(dir);
179 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
181 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
184 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
185 void *data, int flag)
187 struct lustre_handle lockh;
192 case LDLM_CB_BLOCKING:
193 ldlm_lock2handle(lock, &lockh);
194 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
196 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
200 case LDLM_CB_CANCELING: {
201 struct inode *inode = ll_inode_from_resource_lock(lock);
202 __u64 bits = lock->l_policy_data.l_inodebits.bits;
204 /* Inode is set to lock->l_resource->lr_lvb_inode
205 * for mdc - bug 24555 */
206 LASSERT(lock->l_ast_data == NULL);
211 /* Invalidate all dentries associated with this inode */
212 LASSERT(ldlm_is_canceling(lock));
214 if (!fid_res_name_eq(ll_inode2fid(inode),
215 &lock->l_resource->lr_name)) {
216 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
217 PFID(ll_inode2fid(inode)), inode);
221 if (bits & MDS_INODELOCK_XATTR) {
222 if (S_ISDIR(inode->i_mode))
223 ll_i2info(inode)->lli_def_stripe_offset = -1;
224 ll_xattr_cache_destroy(inode);
225 bits &= ~MDS_INODELOCK_XATTR;
228 /* For OPEN locks we differentiate between lock modes
229 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
230 if (bits & MDS_INODELOCK_OPEN)
231 ll_have_md_lock(inode, &bits, lock->l_req_mode);
233 if (bits & MDS_INODELOCK_OPEN) {
236 switch (lock->l_req_mode) {
247 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
251 ll_md_real_close(inode, fmode);
253 bits &= ~MDS_INODELOCK_OPEN;
256 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
257 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
258 ll_have_md_lock(inode, &bits, LCK_MINMODE);
260 if (bits & MDS_INODELOCK_LAYOUT) {
261 struct cl_object_conf conf = {
262 .coc_opc = OBJECT_CONF_INVALIDATE,
266 rc = ll_layout_conf(inode, &conf);
268 CDEBUG(D_INODE, "cannot invalidate layout of "
270 PFID(ll_inode2fid(inode)), rc);
273 if (bits & MDS_INODELOCK_UPDATE) {
274 struct ll_inode_info *lli = ll_i2info(inode);
275 lli->lli_update_atime = 1;
278 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
279 struct ll_inode_info *lli = ll_i2info(inode);
281 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
282 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
283 lli, PFID(&lli->lli_pfid));
284 truncate_inode_pages(inode->i_mapping, 0);
286 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
287 struct inode *master_inode = NULL;
290 /* This is slave inode, since all of the child
291 * dentry is connected on the master inode, so
292 * we have to invalidate the negative children
294 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
295 PFID(ll_inode2fid(inode)),
296 PFID(&lli->lli_pfid));
298 hash = cl_fid_build_ino(&lli->lli_pfid,
299 ll_need_32bit_api(ll_i2sbi(inode)));
301 /* Do not lookup the inode with ilookup5,
302 * otherwise it will cause dead lock,
304 * 1. Client1 send chmod req to the MDT0, then
305 * on MDT0, it enqueues master and all of its
306 * slaves lock, (mdt_attr_set() ->
307 * mdt_lock_slaves()), after gets master and
308 * stripe0 lock, it will send the enqueue req
309 * (for stripe1) to MDT1, then MDT1 finds the
310 * lock has been granted to client2. Then MDT1
311 * sends blocking ast to client2.
313 * 2. At the same time, client2 tries to unlink
314 * the striped dir (rm -rf striped_dir), and
315 * during lookup, it will hold the master inode
316 * of the striped directory, whose inode state
317 * is NEW, then tries to revalidate all of its
318 * slaves, (ll_prep_inode()->ll_iget()->
319 * ll_read_inode2()-> ll_update_inode().). And
320 * it will be blocked on the server side because
323 * 3. Then the client get the blocking_ast req,
324 * cancel the lock, but being blocked if using
325 * ->ilookup5()), because master inode state is
327 master_inode = ilookup5_nowait(inode->i_sb,
328 hash, ll_test_inode_by_fid,
329 (void *)&lli->lli_pfid);
331 ll_invalidate_negative_children(
336 ll_invalidate_negative_children(inode);
340 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
341 inode->i_sb->s_root != NULL &&
342 inode != inode->i_sb->s_root->d_inode)
343 ll_invalidate_aliases(inode);
355 __u32 ll_i2suppgid(struct inode *i)
357 if (in_group_p(i->i_gid))
358 return (__u32)from_kgid(&init_user_ns, i->i_gid);
360 return (__u32) __kgid_val(INVALID_GID);
363 /* Pack the required supplementary groups into the supplied groups array.
364 * If we don't need to use the groups from the target inode(s) then we
365 * instead pack one or more groups from the user's supplementary group
366 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
367 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
370 LASSERT(suppgids != NULL);
372 suppgids[0] = ll_i2suppgid(i1);
375 suppgids[1] = ll_i2suppgid(i2);
381 * try to reuse three types of dentry:
382 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
383 * by concurrent .revalidate).
384 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
385 * be cleared by others calling d_lustre_revalidate).
386 * 3. DISCONNECTED alias.
388 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
390 struct dentry *alias, *discon_alias, *invalid_alias;
391 DECLARE_LL_D_HLIST_NODE_PTR(p);
393 if (ll_d_hlist_empty(&inode->i_dentry))
396 discon_alias = invalid_alias = NULL;
398 ll_lock_dcache(inode);
399 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
400 LASSERT(alias != dentry);
402 spin_lock(&alias->d_lock);
403 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
404 S_ISDIR(inode->i_mode))
405 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
406 discon_alias = alias;
407 else if (alias->d_parent == dentry->d_parent &&
408 alias->d_name.hash == dentry->d_name.hash &&
409 alias->d_name.len == dentry->d_name.len &&
410 memcmp(alias->d_name.name, dentry->d_name.name,
411 dentry->d_name.len) == 0)
412 invalid_alias = alias;
413 spin_unlock(&alias->d_lock);
418 alias = invalid_alias ?: discon_alias ?: NULL;
420 spin_lock(&alias->d_lock);
422 spin_unlock(&alias->d_lock);
424 ll_unlock_dcache(inode);
430 * Similar to d_splice_alias(), but lustre treats invalid alias
431 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
433 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
439 new = ll_find_alias(inode, de);
449 "Reuse dentry %p inode %p refc %d flags %#x\n",
450 new, new->d_inode, ll_d_count(new), new->d_flags);
458 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
459 de, de->d_inode, ll_d_count(de), de->d_flags);
463 static int ll_lookup_it_finish(struct ptlrpc_request *request,
464 struct lookup_intent *it,
465 struct inode *parent, struct dentry **de)
467 struct inode *inode = NULL;
470 struct dentry *alias;
473 /* NB 1 request reference will be taken away by ll_intent_lock()
475 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
477 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
478 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
482 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
484 /* We used to query real size from OSTs here, but actually
485 this is not needed. For stat() calls size would be updated
486 from subsequent do_revalidate()->ll_inode_revalidate_it() in
488 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
489 Everybody else who needs correct file size would call
490 ll_glimpse_size or some equivalent themselves anyway.
491 Also see bug 7198. */
494 /* Only hash *de if it is unhashed (new dentry).
495 * Atoimc_open may passin hashed dentries for open.
497 alias = ll_splice_alias(inode, *de);
499 GOTO(out, rc = PTR_ERR(alias));
503 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
504 /* we have lookup look - unhide dentry */
505 if (bits & MDS_INODELOCK_LOOKUP)
506 d_lustre_revalidate(*de);
507 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
508 /* If file created on server, don't depend on parent UPDATE
509 * lock to unhide it. It is left hidden and next lookup can
510 * find it in ll_splice_alias.
512 /* Check that parent has UPDATE lock. */
513 struct lookup_intent parent_it = {
515 .it_lock_handle = 0 };
516 struct lu_fid fid = ll_i2info(parent)->lli_fid;
518 /* If it is striped directory, get the real stripe parent */
519 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
520 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
521 ll_i2info(parent)->lli_lsm_md,
523 (*de)->d_name.len, &fid);
528 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
530 d_lustre_revalidate(*de);
531 ll_intent_release(&parent_it);
538 if (rc != 0 && it->it_op & IT_OPEN)
539 ll_open_cleanup((*de)->d_sb, request);
544 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
545 struct lookup_intent *it,
546 void **secctx, __u32 *secctxlen)
548 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
549 struct dentry *save = dentry, *retval;
550 struct ptlrpc_request *req = NULL;
551 struct md_op_data *op_data = NULL;
556 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
557 RETURN(ERR_PTR(-ENAMETOOLONG));
559 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
560 dentry->d_name.len, dentry->d_name.name,
561 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
563 if (d_mountpoint(dentry))
564 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
566 if (it == NULL || it->it_op == IT_GETXATTR)
569 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
570 rc = ll_statahead(parent, &dentry, 0);
572 RETURN(dentry == save ? NULL : dentry);
575 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
576 dentry->d_sb->s_flags & MS_RDONLY)
577 RETURN(ERR_PTR(-EROFS));
579 if (it->it_op & IT_CREAT)
580 opc = LUSTRE_OPC_CREATE;
582 opc = LUSTRE_OPC_ANY;
584 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
585 dentry->d_name.len, 0, opc, NULL);
587 GOTO(out, retval = ERR_CAST(op_data));
589 /* enforce umask if acl disabled or MDS doesn't support umask */
590 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
591 it->it_create_mode &= ~current_umask();
593 if (it->it_op & IT_CREAT &&
594 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
595 rc = ll_dentry_init_security(dentry, it->it_create_mode,
597 &op_data->op_file_secctx_name,
598 &op_data->op_file_secctx,
599 &op_data->op_file_secctx_size);
601 GOTO(out, retval = ERR_PTR(rc));
603 *secctx = op_data->op_file_secctx;
604 if (secctxlen != NULL)
605 *secctxlen = op_data->op_file_secctx_size;
608 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
609 &ll_md_blocking_ast, 0);
610 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
611 * client does not know which suppgid should be sent to the MDS, or
612 * some other(s) changed the target file's GID after this RPC sent
613 * to the MDS with the suppgid as the original GID, then we should
614 * try again with right suppgid. */
615 if (rc == -EACCES && it->it_op & IT_OPEN &&
616 it_disposition(it, DISP_OPEN_DENY)) {
617 struct mdt_body *body;
619 LASSERT(req != NULL);
621 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
622 if (op_data->op_suppgids[0] == body->mbo_gid ||
623 op_data->op_suppgids[1] == body->mbo_gid ||
624 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
625 GOTO(out, retval = ERR_PTR(-EACCES));
627 fid_zero(&op_data->op_fid2);
628 op_data->op_suppgids[1] = body->mbo_gid;
629 ptlrpc_req_finished(req);
631 ll_intent_release(it);
632 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
633 &ll_md_blocking_ast, 0);
637 GOTO(out, retval = ERR_PTR(rc));
639 rc = ll_lookup_it_finish(req, it, parent, &dentry);
641 ll_intent_release(it);
642 GOTO(out, retval = ERR_PTR(rc));
645 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
646 !S_ISREG(dentry->d_inode->i_mode) &&
647 !S_ISDIR(dentry->d_inode->i_mode)) {
648 ll_release_openhandle(dentry, it);
650 ll_lookup_finish_locks(it, dentry);
652 GOTO(out, retval = (dentry == save) ? NULL : dentry);
655 if (op_data != NULL && !IS_ERR(op_data)) {
656 if (secctx != NULL && secctxlen != NULL) {
657 /* caller needs sec ctx info, so reset it in op_data to
658 * prevent it from being freed */
659 op_data->op_file_secctx = NULL;
660 op_data->op_file_secctx_size = 0;
662 ll_finish_md_op_data(op_data);
665 ptlrpc_req_finished(req);
669 #ifdef HAVE_IOP_ATOMIC_OPEN
670 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
673 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
676 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
677 dentry->d_name.len, dentry->d_name.name,
678 PFID(ll_inode2fid(parent)), parent, flags);
681 * Optimize away (CREATE && !OPEN). Let .create handle the race.
682 * but only if we have write permissions there, otherwise we need
683 * to proceed with lookup. LU-4185
685 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
686 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
689 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
693 de = ll_lookup_it(parent, dentry, itp, NULL, NULL);
696 ll_intent_release(itp);
702 * For cached negative dentry and new dentry, handle lookup/create/open
705 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
706 struct file *file, unsigned open_flags,
707 umode_t mode, int *opened)
709 struct lookup_intent *it;
711 long long lookup_flags = LOOKUP_OPEN;
717 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
718 "open_flags %x, mode %x opened %d\n",
719 dentry->d_name.len, dentry->d_name.name,
720 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
722 /* Only negative dentries enter here */
723 LASSERT(dentry->d_inode == NULL);
725 if (!d_unhashed(dentry)) {
726 /* A valid negative dentry that just passed revalidation,
727 * there's little point to try and open it server-side,
728 * even though there's a minuscule chance it might succeed.
729 * Either way it's a valid race to just return -ENOENT here.
731 if (!(open_flags & O_CREAT))
734 /* Otherwise we just unhash it to be rehashed afresh via
735 * lookup if necessary
740 OBD_ALLOC(it, sizeof(*it));
745 if (open_flags & O_CREAT) {
746 it->it_op |= IT_CREAT;
747 lookup_flags |= LOOKUP_CREATE;
749 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
750 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
751 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
753 /* Dentry added to dcache tree in ll_lookup_it */
754 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen);
760 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
763 if (it_disposition(it, DISP_OPEN_CREATE)) {
764 /* Dentry instantiated in ll_create_it. */
765 rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
766 security_release_secctx(secctx, secctxlen);
768 /* We dget in ll_splice_alias. */
774 *opened |= FILE_CREATED;
776 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
778 if (S_ISFIFO(dentry->d_inode->i_mode)) {
779 /* We cannot call open here as it might
780 * deadlock. This case is unreachable in
781 * practice because of OBD_CONNECT_NODEVOH. */
782 rc = finish_no_open(file, de);
784 file->private_data = it;
785 rc = finish_open(file, dentry, NULL, opened);
786 /* We dget in ll_splice_alias. finish_open takes
787 * care of dget for fd open.
793 rc = finish_no_open(file, de);
798 ll_intent_release(it);
799 OBD_FREE(it, sizeof(*it));
804 #else /* !HAVE_IOP_ATOMIC_OPEN */
805 static struct lookup_intent *
806 ll_convert_intent(struct open_intent *oit, int lookup_flags)
808 struct lookup_intent *it;
812 return ERR_PTR(-ENOMEM);
814 if (lookup_flags & LOOKUP_OPEN) {
816 if (lookup_flags & LOOKUP_CREATE)
817 it->it_op |= IT_CREAT;
818 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
819 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
820 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
822 it->it_op = IT_GETATTR;
828 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
829 struct nameidata *nd)
834 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
835 struct lookup_intent *it;
837 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
838 it = ll_d2d(dentry)->lld_it;
839 ll_d2d(dentry)->lld_it = NULL;
842 * Optimize away (CREATE && !OPEN). Let .create handle
843 * the race. But only if we have write permissions
844 * there, otherwise we need to proceed with lookup.
847 if ((nd->flags & LOOKUP_CREATE) &&
848 !(nd->flags & LOOKUP_OPEN) &&
849 (inode_permission(parent,
850 MAY_WRITE | MAY_EXEC) == 0))
853 it = ll_convert_intent(&nd->intent.open, nd->flags);
855 RETURN((struct dentry *)it);
858 de = ll_lookup_it(parent, dentry, it, NULL, NULL);
861 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
862 if (dentry->d_inode &&
863 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
864 if (S_ISFIFO(dentry->d_inode->i_mode)) {
865 /* We cannot call open here as it might
866 * deadlock. This case is unreachable in
867 * practice because of
868 * OBD_CONNECT_NODEVOH. */
872 nd->intent.open.file->private_data = it;
873 filp = lookup_instantiate_filp(nd,
879 de = (struct dentry *)filp;
882 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
883 /* XXX This can only reliably work on assumption
884 * that there are NO hashed negative dentries.*/
885 ll_d2d(dentry)->lld_it = it;
886 it = NULL; /* Will be freed in ll_create_nd */
887 /* We absolutely depend on ll_create_nd to be
888 * called to not leak this intent and possible
889 * data attached to it */
894 ll_intent_release(it);
895 OBD_FREE(it, sizeof(*it));
898 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL);
903 #endif /* HAVE_IOP_ATOMIC_OPEN */
905 /* We depend on "mode" being set with the proper file type/umask by now */
906 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
908 struct inode *inode = NULL;
909 struct ptlrpc_request *request = NULL;
910 struct ll_sb_info *sbi = ll_i2sbi(dir);
914 LASSERT(it && it->it_disposition);
916 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
917 request = it->it_request;
918 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
919 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
921 GOTO(out, inode = ERR_PTR(rc));
923 /* Pause to allow for a race with concurrent access by fid */
924 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
926 /* We asked for a lock on the directory, but were granted a
927 * lock on the inode. Since we finally have an inode pointer,
928 * stuff it in the lock. */
929 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
930 PFID(ll_inode2fid(inode)), inode);
931 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
934 ptlrpc_req_finished(request);
939 * By the time this is called, we already have created the directory cache
940 * entry for the new file, but it is so far negative - it has no inode.
942 * We defer creating the OBD object(s) until open, to keep the intent and
943 * non-intent code paths similar, and also because we do not have the MDS
944 * inode number before calling ll_create_node() (which is needed for LOV),
945 * so we would need to do yet another RPC to the MDS to store the LOV EA
946 * data on the MDS. If needed, we would pass the PACKED lmm as data and
947 * lmm_size in datalen (the MDS still has code which will handle that).
949 * If the create succeeds, we fill in the inode information
950 * with d_instantiate().
952 static int ll_create_it(struct inode *dir, struct dentry *dentry,
953 struct lookup_intent *it,
954 void *secctx, __u32 secctxlen)
960 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
961 dentry->d_name.len, dentry->d_name.name,
962 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
964 rc = it_open_error(DISP_OPEN_CREATE, it);
968 inode = ll_create_node(dir, it);
970 RETURN(PTR_ERR(inode));
972 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
975 /* must be done before d_instantiate, because it calls
976 * security_d_instantiate, which means a getxattr if security
977 * context is not set yet */
978 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
984 d_instantiate(dentry, inode);
986 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
987 rc = ll_inode_init_security(dentry, inode, dir);
995 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
997 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1001 if (body->mbo_valid & OBD_MD_FLMTIME &&
1002 body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1003 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to %llu"
1004 "\n", PFID(ll_inode2fid(inode)),
1005 LTIME_S(inode->i_mtime), body->mbo_mtime);
1006 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1009 if (body->mbo_valid & OBD_MD_FLCTIME &&
1010 body->mbo_ctime > LTIME_S(inode->i_ctime))
1011 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1014 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1015 const char *tgt, umode_t mode, int rdev, __u32 opc)
1017 struct qstr *name = &dchild->d_name;
1018 struct ptlrpc_request *request = NULL;
1019 struct md_op_data *op_data;
1020 struct inode *inode = NULL;
1021 struct ll_sb_info *sbi = ll_i2sbi(dir);
1026 if (unlikely(tgt != NULL))
1027 tgt_len = strlen(tgt) + 1;
1030 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1031 name->len, 0, opc, NULL);
1032 if (IS_ERR(op_data))
1033 GOTO(err_exit, err = PTR_ERR(op_data));
1035 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1036 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1037 &op_data->op_file_secctx_name,
1038 &op_data->op_file_secctx,
1039 &op_data->op_file_secctx_size);
1041 GOTO(err_exit, err);
1044 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1045 from_kuid(&init_user_ns, current_fsuid()),
1046 from_kgid(&init_user_ns, current_fsgid()),
1047 cfs_curproc_cap_pack(), rdev, &request);
1048 if (err < 0 && err != -EREMOTE)
1049 GOTO(err_exit, err);
1051 /* If the client doesn't know where to create a subdirectory (or
1052 * in case of a race that sends the RPC to the wrong MDS), the
1053 * MDS will return -EREMOTE and the client will fetch the layout
1054 * of the directory, then create the directory on the right MDT. */
1055 if (unlikely(err == -EREMOTE)) {
1056 struct ll_inode_info *lli = ll_i2info(dir);
1057 struct lmv_user_md *lum;
1061 ptlrpc_req_finished(request);
1064 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1065 OBD_MD_DEFAULT_MEA);
1067 /* Update stripe_offset and retry */
1068 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1069 } else if (err2 == -ENODATA &&
1070 lli->lli_def_stripe_offset != -1) {
1071 /* If there are no default stripe EA on the MDT, but the
1072 * client has default stripe, then it probably means
1073 * default stripe EA has just been deleted. */
1074 lli->lli_def_stripe_offset = -1;
1076 GOTO(err_exit, err);
1079 ptlrpc_req_finished(request);
1081 ll_finish_md_op_data(op_data);
1085 ll_update_times(request, dir);
1087 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1089 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1091 GOTO(err_exit, err);
1093 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1095 /* must be done before d_instantiate, because it calls
1096 * security_d_instantiate, which means a getxattr if security
1097 * context is not set yet */
1098 err = security_inode_notifysecctx(inode,
1099 op_data->op_file_secctx,
1100 op_data->op_file_secctx_size);
1101 inode_unlock(inode);
1103 GOTO(err_exit, err);
1106 d_instantiate(dchild, inode);
1108 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1109 err = ll_inode_init_security(dchild, inode, dir);
1111 GOTO(err_exit, err);
1116 if (request != NULL)
1117 ptlrpc_req_finished(request);
1119 if (!IS_ERR_OR_NULL(op_data))
1120 ll_finish_md_op_data(op_data);
1125 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1128 struct qstr *name = &dchild->d_name;
1132 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1133 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1136 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1137 mode &= ~current_umask();
1139 switch (mode & S_IFMT) {
1141 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1147 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1158 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1163 #ifdef HAVE_IOP_ATOMIC_OPEN
1165 * Plain create. Intent create is handled in atomic_open.
1167 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1168 umode_t mode, bool want_excl)
1172 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1174 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1175 "flags=%u, excl=%d\n", dentry->d_name.len,
1176 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1177 dir, mode, want_excl);
1179 /* Using mknod(2) to create a regular file is designed to not recognize
1180 * volatile file name, so we use ll_mknod() here. */
1181 rc = ll_mknod(dir, dentry, mode, 0);
1183 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1185 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1186 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1190 #else /* !HAVE_IOP_ATOMIC_OPEN */
1191 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1192 ll_umode_t mode, struct nameidata *nd)
1194 struct ll_dentry_data *lld = ll_d2d(dentry);
1195 struct lookup_intent *it = NULL;
1198 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1204 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1205 * so that volatile file name is recoginized.
1206 * Mknod(2), however, is designed to not recognize volatile
1207 * file name to avoid inode leak under orphan directory until
1209 return ll_new_node(dir, dentry, NULL, mode, 0,
1215 /* Was there an error? Propagate it! */
1216 if (it->it_status) {
1221 rc = ll_create_it(dir, dentry, it, NULL, 0);
1222 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1225 nd->intent.open.file->private_data = it;
1226 filp = lookup_instantiate_filp(nd, dentry, NULL);
1232 ll_intent_release(it);
1233 OBD_FREE(it, sizeof(*it));
1236 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1240 #endif /* HAVE_IOP_ATOMIC_OPEN */
1242 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1243 const char *oldpath)
1245 struct qstr *name = &dchild->d_name;
1249 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1250 name->len, name->name, PFID(ll_inode2fid(dir)),
1251 dir, 3000, oldpath);
1253 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1254 LUSTRE_OPC_SYMLINK);
1257 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1262 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1263 struct dentry *new_dentry)
1265 struct inode *src = old_dentry->d_inode;
1266 struct qstr *name = &new_dentry->d_name;
1267 struct ll_sb_info *sbi = ll_i2sbi(dir);
1268 struct ptlrpc_request *request = NULL;
1269 struct md_op_data *op_data;
1273 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1274 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1275 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1277 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1278 0, LUSTRE_OPC_ANY, NULL);
1279 if (IS_ERR(op_data))
1280 RETURN(PTR_ERR(op_data));
1282 err = md_link(sbi->ll_md_exp, op_data, &request);
1283 ll_finish_md_op_data(op_data);
1287 ll_update_times(request, dir);
1288 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1291 ptlrpc_req_finished(request);
1295 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1297 struct qstr *name = &dchild->d_name;
1301 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1302 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1304 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1305 mode &= ~current_umask();
1307 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1309 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1311 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1316 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1318 struct qstr *name = &dchild->d_name;
1319 struct ptlrpc_request *request = NULL;
1320 struct md_op_data *op_data;
1324 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1325 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1327 if (unlikely(d_mountpoint(dchild)))
1330 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1331 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1332 if (IS_ERR(op_data))
1333 RETURN(PTR_ERR(op_data));
1335 if (dchild->d_inode != NULL)
1336 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1338 op_data->op_fid2 = op_data->op_fid3;
1339 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1340 ll_finish_md_op_data(op_data);
1342 ll_update_times(request, dir);
1343 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1346 ptlrpc_req_finished(request);
1353 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1355 struct ptlrpc_request *request = NULL;
1356 struct md_op_data *op_data;
1360 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1361 namelen, name, PFID(ll_inode2fid(dir)), dir);
1363 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1364 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1365 if (IS_ERR(op_data))
1366 RETURN(PTR_ERR(op_data));
1367 op_data->op_cli_flags |= CLI_RM_ENTRY;
1368 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1369 ll_finish_md_op_data(op_data);
1371 ll_update_times(request, dir);
1372 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1375 ptlrpc_req_finished(request);
1379 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1381 struct qstr *name = &dchild->d_name;
1382 struct ptlrpc_request *request = NULL;
1383 struct md_op_data *op_data;
1386 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1387 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1390 * XXX: unlink bind mountpoint maybe call to here,
1391 * just check it as vfs_unlink does.
1393 if (unlikely(d_mountpoint(dchild)))
1396 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1397 LUSTRE_OPC_ANY, NULL);
1398 if (IS_ERR(op_data))
1399 RETURN(PTR_ERR(op_data));
1401 if (dchild->d_inode != NULL)
1402 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1404 op_data->op_fid2 = op_data->op_fid3;
1405 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1406 ll_finish_md_op_data(op_data);
1410 ll_update_times(request, dir);
1411 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1414 ptlrpc_req_finished(request);
1418 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1419 struct inode *tgt, struct dentry *tgt_dchild
1420 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1421 , unsigned int flags
1425 struct qstr *src_name = &src_dchild->d_name;
1426 struct qstr *tgt_name = &tgt_dchild->d_name;
1427 struct ptlrpc_request *request = NULL;
1428 struct ll_sb_info *sbi = ll_i2sbi(src);
1429 struct md_op_data *op_data;
1433 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1438 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1439 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1440 src_name->len, src_name->name,
1441 PFID(ll_inode2fid(src)), src, tgt_name->len,
1442 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1444 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1447 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1448 LUSTRE_OPC_ANY, NULL);
1449 if (IS_ERR(op_data))
1450 RETURN(PTR_ERR(op_data));
1452 if (src_dchild->d_inode != NULL)
1453 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1455 if (tgt_dchild->d_inode != NULL)
1456 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1458 err = md_rename(sbi->ll_md_exp, op_data,
1459 src_name->name, src_name->len,
1460 tgt_name->name, tgt_name->len, &request);
1461 ll_finish_md_op_data(op_data);
1463 ll_update_times(request, src);
1464 ll_update_times(request, tgt);
1465 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1468 ptlrpc_req_finished(request);
1471 d_move(src_dchild, tgt_dchild);
1476 const struct inode_operations ll_dir_inode_operations = {
1478 #ifdef HAVE_IOP_ATOMIC_OPEN
1479 .atomic_open = ll_atomic_open,
1481 .lookup = ll_lookup_nd,
1482 .create = ll_create_nd,
1483 /* We need all these non-raw things for NFSD, to not patch it. */
1484 .unlink = ll_unlink,
1487 .symlink = ll_symlink,
1489 .rename = ll_rename,
1490 .setattr = ll_setattr,
1491 .getattr = ll_getattr,
1492 .permission = ll_inode_permission,
1493 .setxattr = ll_setxattr,
1494 .getxattr = ll_getxattr,
1495 .listxattr = ll_listxattr,
1496 .removexattr = ll_removexattr,
1497 #ifdef HAVE_IOP_GET_ACL
1498 .get_acl = ll_get_acl,
1502 const struct inode_operations ll_special_inode_operations = {
1503 .setattr = ll_setattr,
1504 .getattr = ll_getattr,
1505 .permission = ll_inode_permission,
1506 .setxattr = ll_setxattr,
1507 .getxattr = ll_getxattr,
1508 .listxattr = ll_listxattr,
1509 .removexattr = ll_removexattr,
1510 #ifdef HAVE_IOP_GET_ACL
1511 .get_acl = ll_get_acl,