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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 #include <linux/sched.h>
40 #include <linux/quotaops.h>
41 #include <linux/highmem.h>
42 #include <linux/pagemap.h>
43 #include <linux/security.h>
44 #include <linux/user_namespace.h>
45 #ifdef HAVE_UIDGID_HEADER
46 # include <linux/uidgid.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include <obd_support.h>
52 #include <lustre_fid.h>
53 #include <lustre_dlm.h>
54 #include <lustre_ver.h>
55 #include "llite_internal.h"
57 static int ll_create_it(struct inode *dir, struct dentry *dentry,
58 struct lookup_intent *it);
60 /* called from iget5_locked->find_inode() under inode_lock spinlock */
61 static int ll_test_inode(struct inode *inode, void *opaque)
63 struct ll_inode_info *lli = ll_i2info(inode);
64 struct lustre_md *md = opaque;
66 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
67 CERROR("MDS body missing FID\n");
71 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
77 static int ll_set_inode(struct inode *inode, void *opaque)
79 struct ll_inode_info *lli = ll_i2info(inode);
80 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
82 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
83 CERROR("MDS body missing FID\n");
87 lli->lli_fid = body->mbo_fid1;
88 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
89 CERROR("Can not initialize inode "DFID" without object type: "
91 PFID(&lli->lli_fid), body->mbo_valid);
95 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
96 if (unlikely(inode->i_mode == 0)) {
97 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
108 * Get an inode by inode number(@hash), which is already instantiated by
109 * the intent lookup).
111 struct inode *ll_iget(struct super_block *sb, ino_t hash,
112 struct lustre_md *md)
120 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
122 RETURN(ERR_PTR(-ENOMEM));
124 if (inode->i_state & I_NEW) {
125 rc = ll_read_inode2(inode, md);
126 if (rc == 0 && S_ISREG(inode->i_mode) &&
127 ll_i2info(inode)->lli_clob == NULL)
128 rc = cl_file_inode_init(inode, md);
131 /* Let's clear directory lsm here, otherwise
132 * make_bad_inode() will reset the inode mode
133 * to regular, then ll_clear_inode will not
134 * be able to clear lsm_md */
135 if (S_ISDIR(inode->i_mode))
136 ll_dir_clear_lsm_md(inode);
137 make_bad_inode(inode);
138 unlock_new_inode(inode);
142 unlock_new_inode(inode);
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_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
188 void *data, int flag)
190 struct lustre_handle lockh;
195 case LDLM_CB_BLOCKING:
196 ldlm_lock2handle(lock, &lockh);
197 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
199 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
203 case LDLM_CB_CANCELING: {
204 struct inode *inode = ll_inode_from_resource_lock(lock);
205 __u64 bits = lock->l_policy_data.l_inodebits.bits;
207 /* Inode is set to lock->l_resource->lr_lvb_inode
208 * for mdc - bug 24555 */
209 LASSERT(lock->l_ast_data == NULL);
214 /* Invalidate all dentries associated with this inode */
215 LASSERT(ldlm_is_canceling(lock));
217 if (!fid_res_name_eq(ll_inode2fid(inode),
218 &lock->l_resource->lr_name)) {
219 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
220 PFID(ll_inode2fid(inode)), inode);
224 if (bits & MDS_INODELOCK_XATTR) {
225 if (S_ISDIR(inode->i_mode))
226 ll_i2info(inode)->lli_def_stripe_offset = -1;
227 ll_xattr_cache_destroy(inode);
228 bits &= ~MDS_INODELOCK_XATTR;
231 /* For OPEN locks we differentiate between lock modes
232 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
233 if (bits & MDS_INODELOCK_OPEN)
234 ll_have_md_lock(inode, &bits, lock->l_req_mode);
236 if (bits & MDS_INODELOCK_OPEN) {
239 switch (lock->l_req_mode) {
250 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
254 ll_md_real_close(inode, fmode);
256 bits &= ~MDS_INODELOCK_OPEN;
259 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
260 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
261 ll_have_md_lock(inode, &bits, LCK_MINMODE);
263 if (bits & MDS_INODELOCK_LAYOUT) {
264 struct cl_object_conf conf = {
265 .coc_opc = OBJECT_CONF_INVALIDATE,
269 rc = ll_layout_conf(inode, &conf);
271 CDEBUG(D_INODE, "cannot invalidate layout of "
273 PFID(ll_inode2fid(inode)), rc);
276 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
277 struct ll_inode_info *lli = ll_i2info(inode);
279 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
280 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
281 lli, PFID(&lli->lli_pfid));
282 truncate_inode_pages(inode->i_mapping, 0);
284 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
285 struct inode *master_inode = NULL;
288 /* This is slave inode, since all of the child
289 * dentry is connected on the master inode, so
290 * we have to invalidate the negative children
292 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
293 PFID(ll_inode2fid(inode)),
294 PFID(&lli->lli_pfid));
296 hash = cl_fid_build_ino(&lli->lli_pfid,
297 ll_need_32bit_api(ll_i2sbi(inode)));
299 /* Do not lookup the inode with ilookup5,
300 * otherwise it will cause dead lock,
302 * 1. Client1 send chmod req to the MDT0, then
303 * on MDT0, it enqueues master and all of its
304 * slaves lock, (mdt_attr_set() ->
305 * mdt_lock_slaves()), after gets master and
306 * stripe0 lock, it will send the enqueue req
307 * (for stripe1) to MDT1, then MDT1 finds the
308 * lock has been granted to client2. Then MDT1
309 * sends blocking ast to client2.
311 * 2. At the same time, client2 tries to unlink
312 * the striped dir (rm -rf striped_dir), and
313 * during lookup, it will hold the master inode
314 * of the striped directory, whose inode state
315 * is NEW, then tries to revalidate all of its
316 * slaves, (ll_prep_inode()->ll_iget()->
317 * ll_read_inode2()-> ll_update_inode().). And
318 * it will be blocked on the server side because
321 * 3. Then the client get the blocking_ast req,
322 * cancel the lock, but being blocked if using
323 * ->ilookup5()), because master inode state is
325 master_inode = ilookup5_nowait(inode->i_sb,
326 hash, ll_test_inode_by_fid,
327 (void *)&lli->lli_pfid);
328 if (master_inode != NULL &&
329 !IS_ERR(master_inode)) {
330 ll_invalidate_negative_children(
335 ll_invalidate_negative_children(inode);
339 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
340 inode->i_sb->s_root != NULL &&
341 inode != inode->i_sb->s_root->d_inode)
342 ll_invalidate_aliases(inode);
354 __u32 ll_i2suppgid(struct inode *i)
356 if (in_group_p(i->i_gid))
357 return (__u32)from_kgid(&init_user_ns, i->i_gid);
359 return (__u32) __kgid_val(INVALID_GID);
362 /* Pack the required supplementary groups into the supplied groups array.
363 * If we don't need to use the groups from the target inode(s) then we
364 * instead pack one or more groups from the user's supplementary group
365 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
366 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
369 LASSERT(suppgids != NULL);
371 suppgids[0] = ll_i2suppgid(i1);
374 suppgids[1] = ll_i2suppgid(i2);
380 * try to reuse three types of dentry:
381 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
382 * by concurrent .revalidate).
383 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
384 * be cleared by others calling d_lustre_revalidate).
385 * 3. DISCONNECTED alias.
387 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
389 struct dentry *alias, *discon_alias, *invalid_alias;
390 DECLARE_LL_D_HLIST_NODE_PTR(p);
392 if (ll_d_hlist_empty(&inode->i_dentry))
395 discon_alias = invalid_alias = NULL;
397 ll_lock_dcache(inode);
398 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
399 LASSERT(alias != dentry);
401 spin_lock(&alias->d_lock);
402 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
403 S_ISDIR(inode->i_mode))
404 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
405 discon_alias = alias;
406 else if (alias->d_parent == dentry->d_parent &&
407 alias->d_name.hash == dentry->d_name.hash &&
408 alias->d_name.len == dentry->d_name.len &&
409 memcmp(alias->d_name.name, dentry->d_name.name,
410 dentry->d_name.len) == 0)
411 invalid_alias = alias;
412 spin_unlock(&alias->d_lock);
417 alias = invalid_alias ?: discon_alias ?: NULL;
419 spin_lock(&alias->d_lock);
421 spin_unlock(&alias->d_lock);
423 ll_unlock_dcache(inode);
429 * Similar to d_splice_alias(), but lustre treats invalid alias
430 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
432 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
438 new = ll_find_alias(inode, de);
448 "Reuse dentry %p inode %p refc %d flags %#x\n",
449 new, new->d_inode, ll_d_count(new), new->d_flags);
457 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
458 de, de->d_inode, ll_d_count(de), de->d_flags);
462 static int ll_lookup_it_finish(struct ptlrpc_request *request,
463 struct lookup_intent *it,
464 struct inode *parent, struct dentry **de)
466 struct inode *inode = NULL;
471 /* NB 1 request reference will be taken away by ll_intent_lock()
473 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
475 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
476 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
480 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
482 /* We used to query real size from OSTs here, but actually
483 this is not needed. For stat() calls size would be updated
484 from subsequent do_revalidate()->ll_inode_revalidate_it() in
486 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
487 Everybody else who needs correct file size would call
488 ll_glimpse_size or some equivalent themselves anyway.
489 Also see bug 7198. */
492 /* Only hash *de if it is unhashed (new dentry).
493 * Atoimc_open may passin hashed dentries for open.
495 if (d_unhashed(*de)) {
496 struct dentry *alias;
498 alias = ll_splice_alias(inode, *de);
500 GOTO(out, rc = PTR_ERR(alias));
503 } else if (!it_disposition(it, DISP_LOOKUP_NEG) &&
504 !it_disposition(it, DISP_OPEN_CREATE)) {
505 /* With DISP_OPEN_CREATE dentry will
506 instantiated in ll_create_it. */
507 LASSERT((*de)->d_inode == NULL);
508 d_instantiate(*de, inode);
511 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
512 /* we have lookup look - unhide dentry */
513 if (bits & MDS_INODELOCK_LOOKUP)
514 d_lustre_revalidate(*de);
515 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
516 /* If file created on server, don't depend on parent UPDATE
517 * lock to unhide it. It is left hidden and next lookup can
518 * find it in ll_splice_alias.
520 /* Check that parent has UPDATE lock. */
521 struct lookup_intent parent_it = {
523 .it_lock_handle = 0 };
524 struct lu_fid fid = ll_i2info(parent)->lli_fid;
526 /* If it is striped directory, get the real stripe parent */
527 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
528 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
529 ll_i2info(parent)->lli_lsm_md,
531 (*de)->d_name.len, &fid);
536 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
538 d_lustre_revalidate(*de);
539 ll_intent_release(&parent_it);
546 if (rc != 0 && it->it_op & IT_OPEN)
547 ll_open_cleanup((*de)->d_sb, request);
552 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
553 struct lookup_intent *it)
555 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
556 struct dentry *save = dentry, *retval;
557 struct ptlrpc_request *req = NULL;
558 struct md_op_data *op_data = NULL;
563 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
564 RETURN(ERR_PTR(-ENAMETOOLONG));
566 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
567 dentry->d_name.len, dentry->d_name.name,
568 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
570 if (d_mountpoint(dentry))
571 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
573 if (it == NULL || it->it_op == IT_GETXATTR)
576 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
577 rc = ll_statahead(parent, &dentry, 0);
579 RETURN(dentry == save ? NULL : dentry);
582 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
583 dentry->d_sb->s_flags & MS_RDONLY)
584 RETURN(ERR_PTR(-EROFS));
586 if (it->it_op & IT_CREAT)
587 opc = LUSTRE_OPC_CREATE;
589 opc = LUSTRE_OPC_ANY;
591 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
592 dentry->d_name.len, 0, opc, NULL);
594 GOTO(out, retval = ERR_CAST(op_data));
596 /* enforce umask if acl disabled or MDS doesn't support umask */
597 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
598 it->it_create_mode &= ~current_umask();
600 if (it->it_op & IT_CREAT &&
601 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
602 rc = ll_dentry_init_security(dentry, it->it_create_mode,
604 &op_data->op_file_secctx_name,
605 &op_data->op_file_secctx,
606 &op_data->op_file_secctx_size);
608 GOTO(out, retval = ERR_PTR(rc));
611 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
612 &ll_md_blocking_ast, 0);
613 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
614 * client does not know which suppgid should be sent to the MDS, or
615 * some other(s) changed the target file's GID after this RPC sent
616 * to the MDS with the suppgid as the original GID, then we should
617 * try again with right suppgid. */
618 if (rc == -EACCES && it->it_op & IT_OPEN &&
619 it_disposition(it, DISP_OPEN_DENY)) {
620 struct mdt_body *body;
622 LASSERT(req != NULL);
624 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
625 if (op_data->op_suppgids[0] == body->mbo_gid ||
626 op_data->op_suppgids[1] == body->mbo_gid ||
627 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
628 GOTO(out, retval = ERR_PTR(-EACCES));
630 fid_zero(&op_data->op_fid2);
631 op_data->op_suppgids[1] = body->mbo_gid;
632 ptlrpc_req_finished(req);
634 ll_intent_release(it);
635 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
636 &ll_md_blocking_ast, 0);
640 GOTO(out, retval = ERR_PTR(rc));
642 rc = ll_lookup_it_finish(req, it, parent, &dentry);
644 ll_intent_release(it);
645 GOTO(out, retval = ERR_PTR(rc));
648 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
649 !S_ISREG(dentry->d_inode->i_mode) &&
650 !S_ISDIR(dentry->d_inode->i_mode)) {
651 ll_release_openhandle(dentry, it);
653 ll_lookup_finish_locks(it, dentry);
655 GOTO(out, retval = (dentry == save) ? NULL : dentry);
658 if (op_data != NULL && !IS_ERR(op_data))
659 ll_finish_md_op_data(op_data);
661 ptlrpc_req_finished(req);
665 #ifdef HAVE_IOP_ATOMIC_OPEN
666 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
669 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
672 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
673 dentry->d_name.len, dentry->d_name.name,
674 PFID(ll_inode2fid(parent)), parent, flags);
676 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
677 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
680 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
684 de = ll_lookup_it(parent, dentry, itp);
687 ll_intent_release(itp);
693 * For cached negative dentry and new dentry, handle lookup/create/open
696 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
697 struct file *file, unsigned open_flags,
698 umode_t mode, int *opened)
700 struct lookup_intent *it;
702 long long lookup_flags = LOOKUP_OPEN;
706 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
707 "open_flags %x, mode %x opened %d\n",
708 dentry->d_name.len, dentry->d_name.name,
709 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
711 OBD_ALLOC(it, sizeof(*it));
716 if (open_flags & O_CREAT) {
717 it->it_op |= IT_CREAT;
718 lookup_flags |= LOOKUP_CREATE;
720 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
721 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
722 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
724 /* Dentry added to dcache tree in ll_lookup_it */
725 de = ll_lookup_it(dir, dentry, it);
731 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
734 if (it_disposition(it, DISP_OPEN_CREATE)) {
735 /* Dentry instantiated in ll_create_it. */
736 rc = ll_create_it(dir, dentry, it);
738 /* We dget in ll_splice_alias. */
744 *opened |= FILE_CREATED;
746 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
748 if (S_ISFIFO(dentry->d_inode->i_mode)) {
749 /* We cannot call open here as it might
750 * deadlock. This case is unreachable in
751 * practice because of OBD_CONNECT_NODEVOH. */
752 rc = finish_no_open(file, de);
754 file->private_data = it;
755 rc = finish_open(file, dentry, NULL, opened);
756 /* We dget in ll_splice_alias. finish_open takes
757 * care of dget for fd open.
763 rc = finish_no_open(file, de);
768 ll_intent_release(it);
769 OBD_FREE(it, sizeof(*it));
774 #else /* !HAVE_IOP_ATOMIC_OPEN */
775 static struct lookup_intent *
776 ll_convert_intent(struct open_intent *oit, int lookup_flags)
778 struct lookup_intent *it;
782 return ERR_PTR(-ENOMEM);
784 if (lookup_flags & LOOKUP_OPEN) {
786 if (lookup_flags & LOOKUP_CREATE)
787 it->it_op |= IT_CREAT;
788 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
789 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
790 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
792 it->it_op = IT_GETATTR;
798 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
799 struct nameidata *nd)
804 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
805 struct lookup_intent *it;
807 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
808 it = ll_d2d(dentry)->lld_it;
809 ll_d2d(dentry)->lld_it = NULL;
811 if ((nd->flags & LOOKUP_CREATE) &&
812 !(nd->flags & LOOKUP_OPEN))
815 it = ll_convert_intent(&nd->intent.open, nd->flags);
817 RETURN((struct dentry *)it);
820 de = ll_lookup_it(parent, dentry, it);
823 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
824 if (dentry->d_inode &&
825 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
826 if (S_ISFIFO(dentry->d_inode->i_mode)) {
827 /* We cannot call open here as it might
828 * deadlock. This case is unreachable in
829 * practice because of
830 * OBD_CONNECT_NODEVOH. */
834 nd->intent.open.file->private_data = it;
835 filp = lookup_instantiate_filp(nd,
841 de = (struct dentry *)filp;
844 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
845 // XXX This can only reliably work on assumption
846 // that there are NO hashed negative dentries.
847 ll_d2d(dentry)->lld_it = it;
848 it = NULL; /* Will be freed in ll_create_nd */
849 /* We absolutely depend on ll_create_nd to be
850 * called to not leak this intent and possible
851 * data attached to it */
856 ll_intent_release(it);
857 OBD_FREE(it, sizeof(*it));
860 de = ll_lookup_it(parent, dentry, NULL);
865 #endif /* HAVE_IOP_ATOMIC_OPEN */
867 /* We depend on "mode" being set with the proper file type/umask by now */
868 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
870 struct inode *inode = NULL;
871 struct ptlrpc_request *request = NULL;
872 struct ll_sb_info *sbi = ll_i2sbi(dir);
876 LASSERT(it && it->it_disposition);
878 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
879 request = it->it_request;
880 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
881 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
883 GOTO(out, inode = ERR_PTR(rc));
885 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
887 /* We asked for a lock on the directory, but were granted a
888 * lock on the inode. Since we finally have an inode pointer,
889 * stuff it in the lock. */
890 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
891 PFID(ll_inode2fid(inode)), inode);
892 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
895 ptlrpc_req_finished(request);
900 * By the time this is called, we already have created the directory cache
901 * entry for the new file, but it is so far negative - it has no inode.
903 * We defer creating the OBD object(s) until open, to keep the intent and
904 * non-intent code paths similar, and also because we do not have the MDS
905 * inode number before calling ll_create_node() (which is needed for LOV),
906 * so we would need to do yet another RPC to the MDS to store the LOV EA
907 * data on the MDS. If needed, we would pass the PACKED lmm as data and
908 * lmm_size in datalen (the MDS still has code which will handle that).
910 * If the create succeeds, we fill in the inode information
911 * with d_instantiate().
913 static int ll_create_it(struct inode *dir, struct dentry *dentry,
914 struct lookup_intent *it)
920 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
921 dentry->d_name.len, dentry->d_name.name,
922 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
924 rc = it_open_error(DISP_OPEN_CREATE, it);
928 inode = ll_create_node(dir, it);
930 RETURN(PTR_ERR(inode));
932 d_instantiate(dentry, inode);
934 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
935 rc = ll_inode_init_security(dentry, inode, dir);
943 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
945 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
949 if (body->mbo_valid & OBD_MD_FLMTIME &&
950 body->mbo_mtime > LTIME_S(inode->i_mtime)) {
951 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
952 "\n", PFID(ll_inode2fid(inode)),
953 LTIME_S(inode->i_mtime), body->mbo_mtime);
954 LTIME_S(inode->i_mtime) = body->mbo_mtime;
957 if (body->mbo_valid & OBD_MD_FLCTIME &&
958 body->mbo_ctime > LTIME_S(inode->i_ctime))
959 LTIME_S(inode->i_ctime) = body->mbo_ctime;
962 static int ll_new_node(struct inode *dir, struct dentry *dchild,
963 const char *tgt, umode_t mode, int rdev, __u32 opc)
965 struct qstr *name = &dchild->d_name;
966 struct ptlrpc_request *request = NULL;
967 struct md_op_data *op_data;
968 struct inode *inode = NULL;
969 struct ll_sb_info *sbi = ll_i2sbi(dir);
974 if (unlikely(tgt != NULL))
975 tgt_len = strlen(tgt) + 1;
978 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
979 name->len, 0, opc, NULL);
981 GOTO(err_exit, err = PTR_ERR(op_data));
983 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
984 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
985 &op_data->op_file_secctx_name,
986 &op_data->op_file_secctx,
987 &op_data->op_file_secctx_size);
992 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
993 from_kuid(&init_user_ns, current_fsuid()),
994 from_kgid(&init_user_ns, current_fsgid()),
995 cfs_curproc_cap_pack(), rdev, &request);
996 ll_finish_md_op_data(op_data);
998 if (err < 0 && err != -EREMOTE)
1001 /* If the client doesn't know where to create a subdirectory (or
1002 * in case of a race that sends the RPC to the wrong MDS), the
1003 * MDS will return -EREMOTE and the client will fetch the layout
1004 * of the directory, then create the directory on the right MDT. */
1005 if (unlikely(err == -EREMOTE)) {
1006 struct ll_inode_info *lli = ll_i2info(dir);
1007 struct lmv_user_md *lum;
1011 ptlrpc_req_finished(request);
1014 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1015 OBD_MD_DEFAULT_MEA);
1017 /* Update stripe_offset and retry */
1018 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1019 } else if (err2 == -ENODATA &&
1020 lli->lli_def_stripe_offset != -1) {
1021 /* If there are no default stripe EA on the MDT, but the
1022 * client has default stripe, then it probably means
1023 * default stripe EA has just been deleted. */
1024 lli->lli_def_stripe_offset = -1;
1026 GOTO(err_exit, err);
1029 ptlrpc_req_finished(request);
1034 ll_update_times(request, dir);
1036 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1038 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1040 GOTO(err_exit, err);
1042 d_instantiate(dchild, inode);
1044 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1045 err = ll_inode_init_security(dchild, inode, dir);
1047 GOTO(err_exit, err);
1052 if (request != NULL)
1053 ptlrpc_req_finished(request);
1055 if (!IS_ERR_OR_NULL(op_data))
1056 ll_finish_md_op_data(op_data);
1061 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1064 struct qstr *name = &dchild->d_name;
1068 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1069 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1072 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1073 mode &= ~current_umask();
1075 switch (mode & S_IFMT) {
1077 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1083 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1094 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1099 #ifdef HAVE_IOP_ATOMIC_OPEN
1101 * Plain create. Intent create is handled in atomic_open.
1103 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1104 umode_t mode, bool want_excl)
1108 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1110 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1111 "flags=%u, excl=%d\n", dentry->d_name.len,
1112 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1113 dir, mode, want_excl);
1115 rc = ll_mknod(dir, dentry, mode, 0);
1117 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1119 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1120 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1124 #else /* !HAVE_IOP_ATOMIC_OPEN */
1125 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1126 ll_umode_t mode, struct nameidata *nd)
1128 struct ll_dentry_data *lld = ll_d2d(dentry);
1129 struct lookup_intent *it = NULL;
1132 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1138 return ll_mknod(dir, dentry, mode, 0);
1142 /* Was there an error? Propagate it! */
1143 if (it->it_status) {
1148 rc = ll_create_it(dir, dentry, it);
1149 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1152 nd->intent.open.file->private_data = it;
1153 filp = lookup_instantiate_filp(nd, dentry, NULL);
1159 ll_intent_release(it);
1160 OBD_FREE(it, sizeof(*it));
1163 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1167 #endif /* HAVE_IOP_ATOMIC_OPEN */
1169 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1170 const char *oldpath)
1172 struct qstr *name = &dchild->d_name;
1176 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1177 name->len, name->name, PFID(ll_inode2fid(dir)),
1178 dir, 3000, oldpath);
1180 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1181 LUSTRE_OPC_SYMLINK);
1184 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1189 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1190 struct dentry *new_dentry)
1192 struct inode *src = old_dentry->d_inode;
1193 struct qstr *name = &new_dentry->d_name;
1194 struct ll_sb_info *sbi = ll_i2sbi(dir);
1195 struct ptlrpc_request *request = NULL;
1196 struct md_op_data *op_data;
1200 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1201 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1202 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1204 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1205 0, LUSTRE_OPC_ANY, NULL);
1206 if (IS_ERR(op_data))
1207 RETURN(PTR_ERR(op_data));
1209 err = md_link(sbi->ll_md_exp, op_data, &request);
1210 ll_finish_md_op_data(op_data);
1214 ll_update_times(request, dir);
1215 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1218 ptlrpc_req_finished(request);
1222 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1224 struct qstr *name = &dchild->d_name;
1228 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1229 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1231 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1232 mode &= ~current_umask();
1234 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1236 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1238 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1243 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1245 struct qstr *name = &dchild->d_name;
1246 struct ptlrpc_request *request = NULL;
1247 struct md_op_data *op_data;
1251 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1252 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1254 if (unlikely(d_mountpoint(dchild)))
1257 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1258 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1259 if (IS_ERR(op_data))
1260 RETURN(PTR_ERR(op_data));
1262 if (dchild->d_inode != NULL)
1263 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1265 op_data->op_fid2 = op_data->op_fid3;
1266 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1267 ll_finish_md_op_data(op_data);
1269 ll_update_times(request, dir);
1270 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1273 ptlrpc_req_finished(request);
1280 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1282 struct ptlrpc_request *request = NULL;
1283 struct md_op_data *op_data;
1287 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1288 namelen, name, PFID(ll_inode2fid(dir)), dir);
1290 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1291 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1292 if (IS_ERR(op_data))
1293 RETURN(PTR_ERR(op_data));
1294 op_data->op_cli_flags |= CLI_RM_ENTRY;
1295 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1296 ll_finish_md_op_data(op_data);
1298 ll_update_times(request, dir);
1299 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1302 ptlrpc_req_finished(request);
1306 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1308 struct qstr *name = &dchild->d_name;
1309 struct ptlrpc_request *request = NULL;
1310 struct md_op_data *op_data;
1313 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1314 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1317 * XXX: unlink bind mountpoint maybe call to here,
1318 * just check it as vfs_unlink does.
1320 if (unlikely(d_mountpoint(dchild)))
1323 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1324 LUSTRE_OPC_ANY, NULL);
1325 if (IS_ERR(op_data))
1326 RETURN(PTR_ERR(op_data));
1328 if (dchild->d_inode != NULL)
1329 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1331 op_data->op_fid2 = op_data->op_fid3;
1332 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1333 ll_finish_md_op_data(op_data);
1337 ll_update_times(request, dir);
1338 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1341 ptlrpc_req_finished(request);
1345 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1346 struct inode *tgt, struct dentry *tgt_dchild)
1348 struct qstr *src_name = &src_dchild->d_name;
1349 struct qstr *tgt_name = &tgt_dchild->d_name;
1350 struct ptlrpc_request *request = NULL;
1351 struct ll_sb_info *sbi = ll_i2sbi(src);
1352 struct md_op_data *op_data;
1355 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1356 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1357 src_name->len, src_name->name,
1358 PFID(ll_inode2fid(src)), src, tgt_name->len,
1359 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1361 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1364 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1365 LUSTRE_OPC_ANY, NULL);
1366 if (IS_ERR(op_data))
1367 RETURN(PTR_ERR(op_data));
1369 if (src_dchild->d_inode != NULL)
1370 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1372 if (tgt_dchild->d_inode != NULL)
1373 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1375 err = md_rename(sbi->ll_md_exp, op_data,
1376 src_name->name, src_name->len,
1377 tgt_name->name, tgt_name->len, &request);
1378 ll_finish_md_op_data(op_data);
1380 ll_update_times(request, src);
1381 ll_update_times(request, tgt);
1382 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1385 ptlrpc_req_finished(request);
1388 d_move(src_dchild, tgt_dchild);
1393 const struct inode_operations ll_dir_inode_operations = {
1395 #ifdef HAVE_IOP_ATOMIC_OPEN
1396 .atomic_open = ll_atomic_open,
1398 .lookup = ll_lookup_nd,
1399 .create = ll_create_nd,
1400 /* We need all these non-raw things for NFSD, to not patch it. */
1401 .unlink = ll_unlink,
1404 .symlink = ll_symlink,
1406 .rename = ll_rename,
1407 .setattr = ll_setattr,
1408 .getattr = ll_getattr,
1409 .permission = ll_inode_permission,
1410 .setxattr = ll_setxattr,
1411 .getxattr = ll_getxattr,
1412 .listxattr = ll_listxattr,
1413 .removexattr = ll_removexattr,
1414 #ifdef HAVE_IOP_GET_ACL
1415 .get_acl = ll_get_acl,
1419 const struct inode_operations ll_special_inode_operations = {
1420 .setattr = ll_setattr,
1421 .getattr = ll_getattr,
1422 .permission = ll_inode_permission,
1423 .setxattr = ll_setxattr,
1424 .getxattr = ll_getxattr,
1425 .listxattr = ll_listxattr,
1426 .removexattr = ll_removexattr,
1427 #ifdef HAVE_IOP_GET_ACL
1428 .get_acl = ll_get_acl,