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 inode_has_no_xattr(inode);
140 unlock_new_inode(inode);
142 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
143 rc = ll_update_inode(inode, md);
144 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
145 PFID(&md->body->mbo_fid1), inode, rc);
147 if (S_ISDIR(inode->i_mode))
148 ll_dir_clear_lsm_md(inode);
157 static void ll_invalidate_negative_children(struct inode *dir)
159 struct dentry *dentry, *tmp_subdir;
160 DECLARE_LL_D_HLIST_NODE_PTR(p);
163 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
164 spin_lock(&dentry->d_lock);
165 if (!list_empty(&dentry->d_subdirs)) {
166 struct dentry *child;
168 list_for_each_entry_safe(child, tmp_subdir,
171 if (child->d_inode == NULL)
172 d_lustre_invalidate(child, 1);
175 spin_unlock(&dentry->d_lock);
177 ll_unlock_dcache(dir);
180 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
182 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
185 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
186 void *data, int flag)
188 struct lustre_handle lockh;
193 case LDLM_CB_BLOCKING:
194 ldlm_lock2handle(lock, &lockh);
195 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
197 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
201 case LDLM_CB_CANCELING: {
202 struct inode *inode = ll_inode_from_resource_lock(lock);
203 __u64 bits = lock->l_policy_data.l_inodebits.bits;
205 /* Inode is set to lock->l_resource->lr_lvb_inode
206 * for mdc - bug 24555 */
207 LASSERT(lock->l_ast_data == NULL);
212 /* Invalidate all dentries associated with this inode */
213 LASSERT(ldlm_is_canceling(lock));
215 if (!fid_res_name_eq(ll_inode2fid(inode),
216 &lock->l_resource->lr_name)) {
217 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
218 PFID(ll_inode2fid(inode)), inode);
222 if (bits & MDS_INODELOCK_XATTR) {
223 if (S_ISDIR(inode->i_mode))
224 ll_i2info(inode)->lli_def_stripe_offset = -1;
225 ll_xattr_cache_destroy(inode);
226 bits &= ~MDS_INODELOCK_XATTR;
229 /* For OPEN locks we differentiate between lock modes
230 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
231 if (bits & MDS_INODELOCK_OPEN)
232 ll_have_md_lock(inode, &bits, lock->l_req_mode);
234 if (bits & MDS_INODELOCK_OPEN) {
237 switch (lock->l_req_mode) {
248 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
252 ll_md_real_close(inode, fmode);
254 bits &= ~MDS_INODELOCK_OPEN;
257 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
258 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
259 ll_have_md_lock(inode, &bits, LCK_MINMODE);
261 if (bits & MDS_INODELOCK_LAYOUT) {
262 struct cl_object_conf conf = {
263 .coc_opc = OBJECT_CONF_INVALIDATE,
267 rc = ll_layout_conf(inode, &conf);
269 CDEBUG(D_INODE, "cannot invalidate layout of "
271 PFID(ll_inode2fid(inode)), rc);
274 if (bits & MDS_INODELOCK_UPDATE) {
275 struct ll_inode_info *lli = ll_i2info(inode);
276 lli->lli_update_atime = 1;
279 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
280 struct ll_inode_info *lli = ll_i2info(inode);
282 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
283 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
284 lli, PFID(&lli->lli_pfid));
285 truncate_inode_pages(inode->i_mapping, 0);
287 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
288 struct inode *master_inode = NULL;
291 /* This is slave inode, since all of the child
292 * dentry is connected on the master inode, so
293 * we have to invalidate the negative children
295 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
296 PFID(ll_inode2fid(inode)),
297 PFID(&lli->lli_pfid));
299 hash = cl_fid_build_ino(&lli->lli_pfid,
300 ll_need_32bit_api(ll_i2sbi(inode)));
302 /* Do not lookup the inode with ilookup5,
303 * otherwise it will cause dead lock,
305 * 1. Client1 send chmod req to the MDT0, then
306 * on MDT0, it enqueues master and all of its
307 * slaves lock, (mdt_attr_set() ->
308 * mdt_lock_slaves()), after gets master and
309 * stripe0 lock, it will send the enqueue req
310 * (for stripe1) to MDT1, then MDT1 finds the
311 * lock has been granted to client2. Then MDT1
312 * sends blocking ast to client2.
314 * 2. At the same time, client2 tries to unlink
315 * the striped dir (rm -rf striped_dir), and
316 * during lookup, it will hold the master inode
317 * of the striped directory, whose inode state
318 * is NEW, then tries to revalidate all of its
319 * slaves, (ll_prep_inode()->ll_iget()->
320 * ll_read_inode2()-> ll_update_inode().). And
321 * it will be blocked on the server side because
324 * 3. Then the client get the blocking_ast req,
325 * cancel the lock, but being blocked if using
326 * ->ilookup5()), because master inode state is
328 master_inode = ilookup5_nowait(inode->i_sb,
329 hash, ll_test_inode_by_fid,
330 (void *)&lli->lli_pfid);
332 ll_invalidate_negative_children(
337 ll_invalidate_negative_children(inode);
341 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
342 inode->i_sb->s_root != NULL &&
343 inode != inode->i_sb->s_root->d_inode)
344 ll_invalidate_aliases(inode);
356 __u32 ll_i2suppgid(struct inode *i)
358 if (in_group_p(i->i_gid))
359 return (__u32)from_kgid(&init_user_ns, i->i_gid);
361 return (__u32) __kgid_val(INVALID_GID);
364 /* Pack the required supplementary groups into the supplied groups array.
365 * If we don't need to use the groups from the target inode(s) then we
366 * instead pack one or more groups from the user's supplementary group
367 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
368 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
371 LASSERT(suppgids != NULL);
373 suppgids[0] = ll_i2suppgid(i1);
376 suppgids[1] = ll_i2suppgid(i2);
382 * try to reuse three types of dentry:
383 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
384 * by concurrent .revalidate).
385 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
386 * be cleared by others calling d_lustre_revalidate).
387 * 3. DISCONNECTED alias.
389 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
391 struct dentry *alias, *discon_alias, *invalid_alias;
392 DECLARE_LL_D_HLIST_NODE_PTR(p);
394 if (ll_d_hlist_empty(&inode->i_dentry))
397 discon_alias = invalid_alias = NULL;
399 ll_lock_dcache(inode);
400 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
401 LASSERT(alias != dentry);
403 spin_lock(&alias->d_lock);
404 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
405 S_ISDIR(inode->i_mode))
406 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
407 discon_alias = alias;
408 else if (alias->d_parent == dentry->d_parent &&
409 alias->d_name.hash == dentry->d_name.hash &&
410 alias->d_name.len == dentry->d_name.len &&
411 memcmp(alias->d_name.name, dentry->d_name.name,
412 dentry->d_name.len) == 0)
413 invalid_alias = alias;
414 spin_unlock(&alias->d_lock);
419 alias = invalid_alias ?: discon_alias ?: NULL;
421 spin_lock(&alias->d_lock);
423 spin_unlock(&alias->d_lock);
425 ll_unlock_dcache(inode);
431 * Similar to d_splice_alias(), but lustre treats invalid alias
432 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
434 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
440 new = ll_find_alias(inode, de);
450 "Reuse dentry %p inode %p refc %d flags %#x\n",
451 new, new->d_inode, ll_d_count(new), new->d_flags);
459 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
460 de, de->d_inode, ll_d_count(de), de->d_flags);
464 static int ll_lookup_it_finish(struct ptlrpc_request *request,
465 struct lookup_intent *it,
466 struct inode *parent, struct dentry **de)
468 struct inode *inode = NULL;
471 struct dentry *alias;
474 /* NB 1 request reference will be taken away by ll_intent_lock()
476 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
478 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
479 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
483 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
485 /* We used to query real size from OSTs here, but actually
486 this is not needed. For stat() calls size would be updated
487 from subsequent do_revalidate()->ll_inode_revalidate_it() in
489 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
490 Everybody else who needs correct file size would call
491 ll_glimpse_size or some equivalent themselves anyway.
492 Also see bug 7198. */
495 /* Only hash *de if it is unhashed (new dentry).
496 * Atoimc_open may passin hashed dentries for open.
498 alias = ll_splice_alias(inode, *de);
500 GOTO(out, rc = PTR_ERR(alias));
504 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
505 /* we have lookup look - unhide dentry */
506 if (bits & MDS_INODELOCK_LOOKUP)
507 d_lustre_revalidate(*de);
508 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
509 /* If file created on server, don't depend on parent UPDATE
510 * lock to unhide it. It is left hidden and next lookup can
511 * find it in ll_splice_alias.
513 /* Check that parent has UPDATE lock. */
514 struct lookup_intent parent_it = {
516 .it_lock_handle = 0 };
517 struct lu_fid fid = ll_i2info(parent)->lli_fid;
519 /* If it is striped directory, get the real stripe parent */
520 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
521 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
522 ll_i2info(parent)->lli_lsm_md,
524 (*de)->d_name.len, &fid);
529 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
531 d_lustre_revalidate(*de);
532 ll_intent_release(&parent_it);
539 if (rc != 0 && it->it_op & IT_OPEN)
540 ll_open_cleanup((*de)->d_sb, request);
545 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
546 struct lookup_intent *it,
547 void **secctx, __u32 *secctxlen)
549 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
550 struct dentry *save = dentry, *retval;
551 struct ptlrpc_request *req = NULL;
552 struct md_op_data *op_data = NULL;
557 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
558 RETURN(ERR_PTR(-ENAMETOOLONG));
560 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
561 dentry->d_name.len, dentry->d_name.name,
562 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
564 if (d_mountpoint(dentry))
565 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
567 if (it == NULL || it->it_op == IT_GETXATTR)
570 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
571 rc = ll_statahead(parent, &dentry, 0);
573 RETURN(dentry == save ? NULL : dentry);
576 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
577 dentry->d_sb->s_flags & MS_RDONLY)
578 RETURN(ERR_PTR(-EROFS));
580 if (it->it_op & IT_CREAT)
581 opc = LUSTRE_OPC_CREATE;
583 opc = LUSTRE_OPC_ANY;
585 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
586 dentry->d_name.len, 0, opc, NULL);
588 GOTO(out, retval = ERR_CAST(op_data));
590 /* enforce umask if acl disabled or MDS doesn't support umask */
591 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
592 it->it_create_mode &= ~current_umask();
594 if (it->it_op & IT_CREAT &&
595 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
596 rc = ll_dentry_init_security(dentry, it->it_create_mode,
598 &op_data->op_file_secctx_name,
599 &op_data->op_file_secctx,
600 &op_data->op_file_secctx_size);
602 GOTO(out, retval = ERR_PTR(rc));
604 *secctx = op_data->op_file_secctx;
605 if (secctxlen != NULL)
606 *secctxlen = op_data->op_file_secctx_size;
609 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
610 &ll_md_blocking_ast, 0);
611 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
612 * client does not know which suppgid should be sent to the MDS, or
613 * some other(s) changed the target file's GID after this RPC sent
614 * to the MDS with the suppgid as the original GID, then we should
615 * try again with right suppgid. */
616 if (rc == -EACCES && it->it_op & IT_OPEN &&
617 it_disposition(it, DISP_OPEN_DENY)) {
618 struct mdt_body *body;
620 LASSERT(req != NULL);
622 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
623 if (op_data->op_suppgids[0] == body->mbo_gid ||
624 op_data->op_suppgids[1] == body->mbo_gid ||
625 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
626 GOTO(out, retval = ERR_PTR(-EACCES));
628 fid_zero(&op_data->op_fid2);
629 op_data->op_suppgids[1] = body->mbo_gid;
630 ptlrpc_req_finished(req);
632 ll_intent_release(it);
633 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
634 &ll_md_blocking_ast, 0);
638 GOTO(out, retval = ERR_PTR(rc));
640 rc = ll_lookup_it_finish(req, it, parent, &dentry);
642 ll_intent_release(it);
643 GOTO(out, retval = ERR_PTR(rc));
646 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
647 !S_ISREG(dentry->d_inode->i_mode) &&
648 !S_ISDIR(dentry->d_inode->i_mode)) {
649 ll_release_openhandle(dentry, it);
651 ll_lookup_finish_locks(it, dentry);
653 GOTO(out, retval = (dentry == save) ? NULL : dentry);
656 if (op_data != NULL && !IS_ERR(op_data)) {
657 if (secctx != NULL && secctxlen != NULL) {
658 /* caller needs sec ctx info, so reset it in op_data to
659 * prevent it from being freed */
660 op_data->op_file_secctx = NULL;
661 op_data->op_file_secctx_size = 0;
663 ll_finish_md_op_data(op_data);
666 ptlrpc_req_finished(req);
670 #ifdef HAVE_IOP_ATOMIC_OPEN
671 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
674 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
677 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
678 dentry->d_name.len, dentry->d_name.name,
679 PFID(ll_inode2fid(parent)), parent, flags);
682 * Optimize away (CREATE && !OPEN). Let .create handle the race.
683 * but only if we have write permissions there, otherwise we need
684 * to proceed with lookup. LU-4185
686 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
687 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
690 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
694 de = ll_lookup_it(parent, dentry, itp, NULL, NULL);
697 ll_intent_release(itp);
703 * For cached negative dentry and new dentry, handle lookup/create/open
706 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
707 struct file *file, unsigned open_flags,
708 umode_t mode, int *opened)
710 struct lookup_intent *it;
712 long long lookup_flags = LOOKUP_OPEN;
718 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
719 "open_flags %x, mode %x opened %d\n",
720 dentry->d_name.len, dentry->d_name.name,
721 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
723 /* Only negative dentries enter here */
724 LASSERT(dentry->d_inode == NULL);
726 if (!d_unhashed(dentry)) {
727 /* A valid negative dentry that just passed revalidation,
728 * there's little point to try and open it server-side,
729 * even though there's a minuscule chance it might succeed.
730 * Either way it's a valid race to just return -ENOENT here.
732 if (!(open_flags & O_CREAT))
735 /* Otherwise we just unhash it to be rehashed afresh via
736 * lookup if necessary
741 OBD_ALLOC(it, sizeof(*it));
746 if (open_flags & O_CREAT) {
747 it->it_op |= IT_CREAT;
748 lookup_flags |= LOOKUP_CREATE;
750 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
751 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
752 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
754 /* Dentry added to dcache tree in ll_lookup_it */
755 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen);
761 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
764 if (it_disposition(it, DISP_OPEN_CREATE)) {
765 /* Dentry instantiated in ll_create_it. */
766 rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
767 security_release_secctx(secctx, secctxlen);
769 /* We dget in ll_splice_alias. */
775 *opened |= FILE_CREATED;
777 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
779 if (S_ISFIFO(dentry->d_inode->i_mode)) {
780 /* We cannot call open here as it might
781 * deadlock. This case is unreachable in
782 * practice because of OBD_CONNECT_NODEVOH. */
783 rc = finish_no_open(file, de);
785 file->private_data = it;
786 rc = finish_open(file, dentry, NULL, opened);
787 /* We dget in ll_splice_alias. finish_open takes
788 * care of dget for fd open.
794 rc = finish_no_open(file, de);
799 ll_intent_release(it);
800 OBD_FREE(it, sizeof(*it));
805 #else /* !HAVE_IOP_ATOMIC_OPEN */
806 static struct lookup_intent *
807 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
809 struct lookup_intent *it;
813 return ERR_PTR(-ENOMEM);
815 if (lookup_flags & LOOKUP_OPEN) {
817 /* Avoid file creation for ro bind mount point(is_readonly) */
818 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
819 it->it_op |= IT_CREAT;
820 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
821 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
822 ~(is_readonly ? O_CREAT : 0));
823 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
825 it->it_op = IT_GETATTR;
831 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
832 struct nameidata *nd)
837 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
838 struct lookup_intent *it;
840 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
841 it = ll_d2d(dentry)->lld_it;
842 ll_d2d(dentry)->lld_it = NULL;
845 * Optimize away (CREATE && !OPEN). Let .create handle
846 * the race. But only if we have write permissions
847 * there, otherwise we need to proceed with lookup.
850 if ((nd->flags & LOOKUP_CREATE) &&
851 !(nd->flags & LOOKUP_OPEN) &&
852 (inode_permission(parent,
853 MAY_WRITE | MAY_EXEC) == 0))
856 it = ll_convert_intent(&nd->intent.open, nd->flags,
857 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
858 (nd->path.mnt->mnt_sb->s_flags & MS_RDONLY));
860 RETURN((struct dentry *)it);
863 de = ll_lookup_it(parent, dentry, it, NULL, NULL);
866 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
867 if (dentry->d_inode &&
868 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
869 if (S_ISFIFO(dentry->d_inode->i_mode)) {
870 /* We cannot call open here as it might
871 * deadlock. This case is unreachable in
872 * practice because of
873 * OBD_CONNECT_NODEVOH. */
877 nd->intent.open.file->private_data = it;
878 filp = lookup_instantiate_filp(nd,
884 de = (struct dentry *)filp;
887 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
888 /* XXX This can only reliably work on assumption
889 * that there are NO hashed negative dentries.*/
890 ll_d2d(dentry)->lld_it = it;
891 it = NULL; /* Will be freed in ll_create_nd */
892 /* We absolutely depend on ll_create_nd to be
893 * called to not leak this intent and possible
894 * data attached to it */
899 ll_intent_release(it);
900 OBD_FREE(it, sizeof(*it));
903 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL);
908 #endif /* HAVE_IOP_ATOMIC_OPEN */
910 /* We depend on "mode" being set with the proper file type/umask by now */
911 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
913 struct inode *inode = NULL;
914 struct ptlrpc_request *request = NULL;
915 struct ll_sb_info *sbi = ll_i2sbi(dir);
919 LASSERT(it && it->it_disposition);
921 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
922 request = it->it_request;
923 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
924 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
926 GOTO(out, inode = ERR_PTR(rc));
928 /* Pause to allow for a race with concurrent access by fid */
929 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
931 /* We asked for a lock on the directory, but were granted a
932 * lock on the inode. Since we finally have an inode pointer,
933 * stuff it in the lock. */
934 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
935 PFID(ll_inode2fid(inode)), inode);
936 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
939 ptlrpc_req_finished(request);
944 * By the time this is called, we already have created the directory cache
945 * entry for the new file, but it is so far negative - it has no inode.
947 * We defer creating the OBD object(s) until open, to keep the intent and
948 * non-intent code paths similar, and also because we do not have the MDS
949 * inode number before calling ll_create_node() (which is needed for LOV),
950 * so we would need to do yet another RPC to the MDS to store the LOV EA
951 * data on the MDS. If needed, we would pass the PACKED lmm as data and
952 * lmm_size in datalen (the MDS still has code which will handle that).
954 * If the create succeeds, we fill in the inode information
955 * with d_instantiate().
957 static int ll_create_it(struct inode *dir, struct dentry *dentry,
958 struct lookup_intent *it,
959 void *secctx, __u32 secctxlen)
965 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
966 dentry->d_name.len, dentry->d_name.name,
967 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
969 rc = it_open_error(DISP_OPEN_CREATE, it);
973 inode = ll_create_node(dir, it);
975 RETURN(PTR_ERR(inode));
977 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
980 /* must be done before d_instantiate, because it calls
981 * security_d_instantiate, which means a getxattr if security
982 * context is not set yet */
983 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
989 d_instantiate(dentry, inode);
991 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
992 rc = ll_inode_init_security(dentry, inode, dir);
1000 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1002 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1006 if (body->mbo_valid & OBD_MD_FLMTIME &&
1007 body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1008 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to %llu"
1009 "\n", PFID(ll_inode2fid(inode)),
1010 LTIME_S(inode->i_mtime), body->mbo_mtime);
1011 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1014 if (body->mbo_valid & OBD_MD_FLCTIME &&
1015 body->mbo_ctime > LTIME_S(inode->i_ctime))
1016 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1019 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1020 const char *tgt, umode_t mode, int rdev, __u32 opc)
1022 struct qstr *name = &dchild->d_name;
1023 struct ptlrpc_request *request = NULL;
1024 struct md_op_data *op_data;
1025 struct inode *inode = NULL;
1026 struct ll_sb_info *sbi = ll_i2sbi(dir);
1031 if (unlikely(tgt != NULL))
1032 tgt_len = strlen(tgt) + 1;
1035 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1036 name->len, 0, opc, NULL);
1037 if (IS_ERR(op_data))
1038 GOTO(err_exit, err = PTR_ERR(op_data));
1040 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1041 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1042 &op_data->op_file_secctx_name,
1043 &op_data->op_file_secctx,
1044 &op_data->op_file_secctx_size);
1046 GOTO(err_exit, err);
1049 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1050 from_kuid(&init_user_ns, current_fsuid()),
1051 from_kgid(&init_user_ns, current_fsgid()),
1052 cfs_curproc_cap_pack(), rdev, &request);
1053 if (err < 0 && err != -EREMOTE)
1054 GOTO(err_exit, err);
1056 /* If the client doesn't know where to create a subdirectory (or
1057 * in case of a race that sends the RPC to the wrong MDS), the
1058 * MDS will return -EREMOTE and the client will fetch the layout
1059 * of the directory, then create the directory on the right MDT. */
1060 if (unlikely(err == -EREMOTE)) {
1061 struct ll_inode_info *lli = ll_i2info(dir);
1062 struct lmv_user_md *lum;
1066 ptlrpc_req_finished(request);
1069 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1070 OBD_MD_DEFAULT_MEA);
1072 /* Update stripe_offset and retry */
1073 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1074 } else if (err2 == -ENODATA &&
1075 lli->lli_def_stripe_offset != -1) {
1076 /* If there are no default stripe EA on the MDT, but the
1077 * client has default stripe, then it probably means
1078 * default stripe EA has just been deleted. */
1079 lli->lli_def_stripe_offset = -1;
1081 GOTO(err_exit, err);
1084 ptlrpc_req_finished(request);
1086 ll_finish_md_op_data(op_data);
1090 ll_update_times(request, dir);
1092 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1094 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1096 GOTO(err_exit, err);
1098 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1100 /* must be done before d_instantiate, because it calls
1101 * security_d_instantiate, which means a getxattr if security
1102 * context is not set yet */
1103 err = security_inode_notifysecctx(inode,
1104 op_data->op_file_secctx,
1105 op_data->op_file_secctx_size);
1106 inode_unlock(inode);
1108 GOTO(err_exit, err);
1111 d_instantiate(dchild, inode);
1113 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1114 err = ll_inode_init_security(dchild, inode, dir);
1116 GOTO(err_exit, err);
1121 if (request != NULL)
1122 ptlrpc_req_finished(request);
1124 if (!IS_ERR_OR_NULL(op_data))
1125 ll_finish_md_op_data(op_data);
1130 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1133 struct qstr *name = &dchild->d_name;
1137 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1138 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1141 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1142 mode &= ~current_umask();
1144 switch (mode & S_IFMT) {
1146 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1152 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1163 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1168 #ifdef HAVE_IOP_ATOMIC_OPEN
1170 * Plain create. Intent create is handled in atomic_open.
1172 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1173 umode_t mode, bool want_excl)
1177 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1179 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1180 "flags=%u, excl=%d\n", dentry->d_name.len,
1181 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1182 dir, mode, want_excl);
1184 /* Using mknod(2) to create a regular file is designed to not recognize
1185 * volatile file name, so we use ll_mknod() here. */
1186 rc = ll_mknod(dir, dentry, mode, 0);
1188 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1190 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1191 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1195 #else /* !HAVE_IOP_ATOMIC_OPEN */
1196 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1197 ll_umode_t mode, struct nameidata *nd)
1199 struct ll_dentry_data *lld = ll_d2d(dentry);
1200 struct lookup_intent *it = NULL;
1203 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1209 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1210 * so that volatile file name is recoginized.
1211 * Mknod(2), however, is designed to not recognize volatile
1212 * file name to avoid inode leak under orphan directory until
1214 return ll_new_node(dir, dentry, NULL, mode, 0,
1220 /* Was there an error? Propagate it! */
1221 if (it->it_status) {
1226 rc = ll_create_it(dir, dentry, it, NULL, 0);
1227 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1230 nd->intent.open.file->private_data = it;
1231 filp = lookup_instantiate_filp(nd, dentry, NULL);
1237 ll_intent_release(it);
1238 OBD_FREE(it, sizeof(*it));
1241 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1245 #endif /* HAVE_IOP_ATOMIC_OPEN */
1247 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1248 const char *oldpath)
1250 struct qstr *name = &dchild->d_name;
1254 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1255 name->len, name->name, PFID(ll_inode2fid(dir)),
1256 dir, 3000, oldpath);
1258 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1259 LUSTRE_OPC_SYMLINK);
1262 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1267 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1268 struct dentry *new_dentry)
1270 struct inode *src = old_dentry->d_inode;
1271 struct qstr *name = &new_dentry->d_name;
1272 struct ll_sb_info *sbi = ll_i2sbi(dir);
1273 struct ptlrpc_request *request = NULL;
1274 struct md_op_data *op_data;
1278 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1279 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1280 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1282 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1283 0, LUSTRE_OPC_ANY, NULL);
1284 if (IS_ERR(op_data))
1285 RETURN(PTR_ERR(op_data));
1287 err = md_link(sbi->ll_md_exp, op_data, &request);
1288 ll_finish_md_op_data(op_data);
1292 ll_update_times(request, dir);
1293 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1296 ptlrpc_req_finished(request);
1300 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1302 struct qstr *name = &dchild->d_name;
1306 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1307 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1309 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1310 mode &= ~current_umask();
1312 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1314 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1316 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1321 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1323 struct qstr *name = &dchild->d_name;
1324 struct ptlrpc_request *request = NULL;
1325 struct md_op_data *op_data;
1329 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1330 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1332 if (unlikely(d_mountpoint(dchild)))
1335 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1336 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1337 if (IS_ERR(op_data))
1338 RETURN(PTR_ERR(op_data));
1340 if (dchild->d_inode != NULL)
1341 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1343 op_data->op_fid2 = op_data->op_fid3;
1344 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1345 ll_finish_md_op_data(op_data);
1347 ll_update_times(request, dir);
1348 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1351 ptlrpc_req_finished(request);
1358 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1360 struct ptlrpc_request *request = NULL;
1361 struct md_op_data *op_data;
1365 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1366 namelen, name, PFID(ll_inode2fid(dir)), dir);
1368 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1369 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1370 if (IS_ERR(op_data))
1371 RETURN(PTR_ERR(op_data));
1372 op_data->op_cli_flags |= CLI_RM_ENTRY;
1373 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1374 ll_finish_md_op_data(op_data);
1376 ll_update_times(request, dir);
1377 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1380 ptlrpc_req_finished(request);
1384 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1386 struct qstr *name = &dchild->d_name;
1387 struct ptlrpc_request *request = NULL;
1388 struct md_op_data *op_data;
1391 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1392 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1395 * XXX: unlink bind mountpoint maybe call to here,
1396 * just check it as vfs_unlink does.
1398 if (unlikely(d_mountpoint(dchild)))
1401 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1402 LUSTRE_OPC_ANY, NULL);
1403 if (IS_ERR(op_data))
1404 RETURN(PTR_ERR(op_data));
1406 if (dchild->d_inode != NULL)
1407 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1409 op_data->op_fid2 = op_data->op_fid3;
1410 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1411 ll_finish_md_op_data(op_data);
1415 ll_update_times(request, dir);
1416 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1419 ptlrpc_req_finished(request);
1423 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1424 struct inode *tgt, struct dentry *tgt_dchild
1425 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1426 , unsigned int flags
1430 struct qstr *src_name = &src_dchild->d_name;
1431 struct qstr *tgt_name = &tgt_dchild->d_name;
1432 struct ptlrpc_request *request = NULL;
1433 struct ll_sb_info *sbi = ll_i2sbi(src);
1434 struct md_op_data *op_data;
1438 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1443 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1444 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1445 src_name->len, src_name->name,
1446 PFID(ll_inode2fid(src)), src, tgt_name->len,
1447 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1449 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1452 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1453 LUSTRE_OPC_ANY, NULL);
1454 if (IS_ERR(op_data))
1455 RETURN(PTR_ERR(op_data));
1457 if (src_dchild->d_inode != NULL)
1458 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1460 if (tgt_dchild->d_inode != NULL)
1461 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1463 err = md_rename(sbi->ll_md_exp, op_data,
1464 src_name->name, src_name->len,
1465 tgt_name->name, tgt_name->len, &request);
1466 ll_finish_md_op_data(op_data);
1468 ll_update_times(request, src);
1469 ll_update_times(request, tgt);
1470 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1473 ptlrpc_req_finished(request);
1476 d_move(src_dchild, tgt_dchild);
1481 const struct inode_operations ll_dir_inode_operations = {
1483 #ifdef HAVE_IOP_ATOMIC_OPEN
1484 .atomic_open = ll_atomic_open,
1486 .lookup = ll_lookup_nd,
1487 .create = ll_create_nd,
1488 /* We need all these non-raw things for NFSD, to not patch it. */
1489 .unlink = ll_unlink,
1492 .symlink = ll_symlink,
1494 .rename = ll_rename,
1495 .setattr = ll_setattr,
1496 .getattr = ll_getattr,
1497 .permission = ll_inode_permission,
1498 .setxattr = ll_setxattr,
1499 .getxattr = ll_getxattr,
1500 .listxattr = ll_listxattr,
1501 .removexattr = ll_removexattr,
1502 #ifdef HAVE_IOP_GET_ACL
1503 .get_acl = ll_get_acl,
1507 const struct inode_operations ll_special_inode_operations = {
1508 .setattr = ll_setattr,
1509 .getattr = ll_getattr,
1510 .permission = ll_inode_permission,
1511 .setxattr = ll_setxattr,
1512 .getxattr = ll_getxattr,
1513 .listxattr = ll_listxattr,
1514 .removexattr = ll_removexattr,
1515 #ifdef HAVE_IOP_GET_ACL
1516 .get_acl = ll_get_acl,