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, 2013, 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>
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);
56 /* called from iget5_locked->find_inode() under inode_lock spinlock */
57 static int ll_test_inode(struct inode *inode, void *opaque)
59 struct ll_inode_info *lli = ll_i2info(inode);
60 struct lustre_md *md = opaque;
62 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
63 CERROR("MDS body missing FID\n");
67 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
73 static int ll_set_inode(struct inode *inode, void *opaque)
75 struct ll_inode_info *lli = ll_i2info(inode);
76 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
78 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
79 CERROR("MDS body missing FID\n");
83 lli->lli_fid = body->mbo_fid1;
84 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
85 CERROR("Can not initialize inode "DFID" without object type: "
87 PFID(&lli->lli_fid), body->mbo_valid);
91 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
92 if (unlikely(inode->i_mode == 0)) {
93 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
104 * Get an inode by inode number(@hash), which is already instantiated by
105 * the intent lookup).
107 struct inode *ll_iget(struct super_block *sb, ino_t hash,
108 struct lustre_md *md)
116 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
118 RETURN(ERR_PTR(-ENOMEM));
120 if (inode->i_state & I_NEW) {
121 rc = ll_read_inode2(inode, md);
122 if (rc == 0 && S_ISREG(inode->i_mode) &&
123 ll_i2info(inode)->lli_clob == NULL) {
124 CDEBUG(D_INODE, "%s: apply lsm %p to inode "DFID"\n",
125 ll_get_fsname(sb, NULL, 0), md->lsm,
126 PFID(ll_inode2fid(inode)));
127 rc = cl_file_inode_init(inode, md);
130 make_bad_inode(inode);
131 unlock_new_inode(inode);
135 unlock_new_inode(inode);
137 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
138 rc = ll_update_inode(inode, md);
139 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
140 PFID(&md->body->mbo_fid1), inode, rc);
150 static void ll_invalidate_negative_children(struct inode *dir)
152 struct dentry *dentry, *tmp_subdir;
153 DECLARE_LL_D_HLIST_NODE_PTR(p);
156 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
157 spin_lock(&dentry->d_lock);
158 if (!list_empty(&dentry->d_subdirs)) {
159 struct dentry *child;
161 list_for_each_entry_safe(child, tmp_subdir,
164 if (child->d_inode == NULL)
165 d_lustre_invalidate(child, 1);
168 spin_unlock(&dentry->d_lock);
170 ll_unlock_dcache(dir);
173 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
175 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
178 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
179 void *data, int flag)
181 struct lustre_handle lockh;
186 case LDLM_CB_BLOCKING:
187 ldlm_lock2handle(lock, &lockh);
188 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
190 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
194 case LDLM_CB_CANCELING: {
195 struct inode *inode = ll_inode_from_resource_lock(lock);
196 __u64 bits = lock->l_policy_data.l_inodebits.bits;
198 /* Inode is set to lock->l_resource->lr_lvb_inode
199 * for mdc - bug 24555 */
200 LASSERT(lock->l_ast_data == NULL);
205 /* Invalidate all dentries associated with this inode */
206 LASSERT(ldlm_is_canceling(lock));
208 if (!fid_res_name_eq(ll_inode2fid(inode),
209 &lock->l_resource->lr_name)) {
210 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
211 PFID(ll_inode2fid(inode)), inode);
215 if (bits & MDS_INODELOCK_XATTR) {
216 ll_xattr_cache_destroy(inode);
217 bits &= ~MDS_INODELOCK_XATTR;
220 /* For OPEN locks we differentiate between lock modes
221 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
222 if (bits & MDS_INODELOCK_OPEN)
223 ll_have_md_lock(inode, &bits, lock->l_req_mode);
225 if (bits & MDS_INODELOCK_OPEN) {
228 switch (lock->l_req_mode) {
239 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
243 ll_md_real_close(inode, fmode);
245 bits &= ~MDS_INODELOCK_OPEN;
248 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
249 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
250 ll_have_md_lock(inode, &bits, LCK_MINMODE);
252 if (bits & MDS_INODELOCK_LAYOUT) {
253 struct cl_object_conf conf = {
254 .coc_opc = OBJECT_CONF_INVALIDATE,
258 rc = ll_layout_conf(inode, &conf);
260 CDEBUG(D_INODE, "cannot invalidate layout of "
262 PFID(ll_inode2fid(inode)), rc);
265 if (bits & MDS_INODELOCK_UPDATE) {
266 struct ll_inode_info *lli = ll_i2info(inode);
268 spin_lock(&lli->lli_lock);
269 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
270 spin_unlock(&lli->lli_lock);
273 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
274 struct ll_inode_info *lli = ll_i2info(inode);
276 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
277 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
278 lli, PFID(&lli->lli_pfid));
279 truncate_inode_pages(inode->i_mapping, 0);
281 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
282 struct inode *master_inode = NULL;
285 /* This is slave inode, since all of the child
286 * dentry is connected on the master inode, so
287 * we have to invalidate the negative children
289 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
290 PFID(ll_inode2fid(inode)),
291 PFID(&lli->lli_pfid));
293 hash = cl_fid_build_ino(&lli->lli_pfid,
294 ll_need_32bit_api(ll_i2sbi(inode)));
296 master_inode = ilookup5(inode->i_sb, hash,
297 ll_test_inode_by_fid,
298 (void *)&lli->lli_pfid);
299 if (master_inode != NULL &&
300 !IS_ERR(master_inode)) {
301 ll_invalidate_negative_children(
306 ll_invalidate_negative_children(inode);
310 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
311 inode->i_sb->s_root != NULL &&
312 inode != inode->i_sb->s_root->d_inode)
313 ll_invalidate_aliases(inode);
325 __u32 ll_i2suppgid(struct inode *i)
327 if (in_group_p(i->i_gid))
328 return (__u32)from_kgid(&init_user_ns, i->i_gid);
330 return (__u32) __kgid_val(INVALID_GID);
333 /* Pack the required supplementary groups into the supplied groups array.
334 * If we don't need to use the groups from the target inode(s) then we
335 * instead pack one or more groups from the user's supplementary group
336 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
337 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
340 LASSERT(suppgids != NULL);
342 suppgids[0] = ll_i2suppgid(i1);
345 suppgids[1] = ll_i2suppgid(i2);
351 * try to reuse three types of dentry:
352 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
353 * by concurrent .revalidate).
354 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
355 * be cleared by others calling d_lustre_revalidate).
356 * 3. DISCONNECTED alias.
358 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
360 struct dentry *alias, *discon_alias, *invalid_alias;
361 DECLARE_LL_D_HLIST_NODE_PTR(p);
363 if (ll_d_hlist_empty(&inode->i_dentry))
366 discon_alias = invalid_alias = NULL;
368 ll_lock_dcache(inode);
369 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
370 LASSERT(alias != dentry);
372 spin_lock(&alias->d_lock);
373 if (alias->d_flags & DCACHE_DISCONNECTED)
374 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
375 discon_alias = alias;
376 else if (alias->d_parent == dentry->d_parent &&
377 alias->d_name.hash == dentry->d_name.hash &&
378 alias->d_name.len == dentry->d_name.len &&
379 memcmp(alias->d_name.name, dentry->d_name.name,
380 dentry->d_name.len) == 0)
381 invalid_alias = alias;
382 spin_unlock(&alias->d_lock);
387 alias = invalid_alias ?: discon_alias ?: NULL;
389 spin_lock(&alias->d_lock);
391 spin_unlock(&alias->d_lock);
393 ll_unlock_dcache(inode);
399 * Similar to d_splice_alias(), but lustre treats invalid alias
400 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
402 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
408 new = ll_find_alias(inode, de);
418 "Reuse dentry %p inode %p refc %d flags %#x\n",
419 new, new->d_inode, ll_d_count(new), new->d_flags);
427 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
428 de, de->d_inode, ll_d_count(de), de->d_flags);
432 static int ll_lookup_it_finish(struct ptlrpc_request *request,
433 struct lookup_intent *it,
434 struct inode *parent, struct dentry **de)
436 struct inode *inode = NULL;
441 /* NB 1 request reference will be taken away by ll_intent_lock()
443 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
444 it->d.lustre.it_disposition);
445 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
446 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
450 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
452 /* We used to query real size from OSTs here, but actually
453 this is not needed. For stat() calls size would be updated
454 from subsequent do_revalidate()->ll_inode_revalidate_it() in
456 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
457 Everybody else who needs correct file size would call
458 ll_glimpse_size or some equivalent themselves anyway.
459 Also see bug 7198. */
462 /* Only hash *de if it is unhashed (new dentry).
463 * Atoimc_open may passin hashed dentries for open.
465 if (d_unhashed(*de)) {
466 struct dentry *alias;
468 alias = ll_splice_alias(inode, *de);
470 RETURN(PTR_ERR(alias));
472 } else if (!it_disposition(it, DISP_LOOKUP_NEG) &&
473 !it_disposition(it, DISP_OPEN_CREATE)) {
474 /* With DISP_OPEN_CREATE dentry will
475 instantiated in ll_create_it. */
476 LASSERT((*de)->d_inode == NULL);
477 d_instantiate(*de, inode);
480 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
481 /* we have lookup look - unhide dentry */
482 if (bits & MDS_INODELOCK_LOOKUP)
483 d_lustre_revalidate(*de);
484 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
485 /* If file created on server, don't depend on parent UPDATE
486 * lock to unhide it. It is left hidden and next lookup can
487 * find it in ll_splice_alias.
489 /* Check that parent has UPDATE lock. */
490 struct lookup_intent parent_it = {
492 .d.lustre.it_lock_handle = 0 };
493 struct lu_fid fid = ll_i2info(parent)->lli_fid;
495 /* If it is striped directory, get the real stripe parent */
496 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
497 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
498 ll_i2info(parent)->lli_lsm_md,
500 (*de)->d_name.len, &fid);
505 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
507 d_lustre_revalidate(*de);
508 ll_intent_release(&parent_it);
515 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
516 struct lookup_intent *it)
518 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
519 struct dentry *save = dentry, *retval;
520 struct ptlrpc_request *req = NULL;
521 struct md_op_data *op_data;
526 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
527 RETURN(ERR_PTR(-ENAMETOOLONG));
529 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
530 dentry->d_name.len, dentry->d_name.name,
531 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
533 if (d_mountpoint(dentry))
534 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
536 if (it == NULL || it->it_op == IT_GETXATTR)
539 if (it->it_op == IT_GETATTR) {
540 rc = ll_statahead_enter(parent, &dentry, 0);
543 GOTO(out, retval = NULL);
544 GOTO(out, retval = dentry);
548 if (it->it_op & IT_CREAT)
549 opc = LUSTRE_OPC_CREATE;
551 opc = LUSTRE_OPC_ANY;
553 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
554 dentry->d_name.len, 0, opc, NULL);
556 RETURN((void *)op_data);
558 /* enforce umask if acl disabled or MDS doesn't support umask */
559 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
560 it->it_create_mode &= ~current_umask();
562 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
563 &ll_md_blocking_ast, 0);
564 ll_finish_md_op_data(op_data);
566 GOTO(out, retval = ERR_PTR(rc));
568 rc = ll_lookup_it_finish(req, it, parent, &dentry);
570 ll_intent_release(it);
571 GOTO(out, retval = ERR_PTR(rc));
574 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
575 !S_ISREG(dentry->d_inode->i_mode) &&
576 !S_ISDIR(dentry->d_inode->i_mode)) {
577 ll_release_openhandle(dentry, it);
579 ll_lookup_finish_locks(it, dentry);
582 GOTO(out, retval = NULL);
584 GOTO(out, retval = dentry);
587 ptlrpc_req_finished(req);
588 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
589 ll_statahead_mark(parent, dentry);
593 #ifdef HAVE_IOP_ATOMIC_OPEN
594 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
597 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
600 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
601 dentry->d_name.len, dentry->d_name.name,
602 PFID(ll_inode2fid(parent)), parent, flags);
604 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
605 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
608 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
612 de = ll_lookup_it(parent, dentry, itp);
615 ll_intent_release(itp);
621 * For cached negative dentry and new dentry, handle lookup/create/open
624 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
625 struct file *file, unsigned open_flags,
626 umode_t mode, int *opened)
628 struct lookup_intent *it;
630 long long lookup_flags = LOOKUP_OPEN;
634 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
635 "open_flags %x, mode %x opened %d\n",
636 dentry->d_name.len, dentry->d_name.name,
637 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
639 OBD_ALLOC(it, sizeof(*it));
644 if (open_flags & O_CREAT) {
645 it->it_op |= IT_CREAT;
646 lookup_flags |= LOOKUP_CREATE;
648 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
649 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
650 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
652 /* Dentry added to dcache tree in ll_lookup_it */
653 de = ll_lookup_it(dir, dentry, it);
660 if (it_disposition(it, DISP_OPEN_CREATE)) {
661 /* Dentry instantiated in ll_create_it. */
662 rc = ll_create_it(dir, dentry, it);
664 /* We dget in ll_splice_alias. */
670 *opened |= FILE_CREATED;
672 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
674 if (S_ISFIFO(dentry->d_inode->i_mode)) {
675 /* We cannot call open here as it would
678 if (it_disposition(it, DISP_ENQ_OPEN_REF))
680 (struct ptlrpc_request *)
681 it->d.lustre.it_data);
682 rc = finish_no_open(file, de);
684 file->private_data = it;
685 rc = finish_open(file, dentry, NULL, opened);
686 /* We dget in ll_splice_alias. finish_open takes
687 * care of dget for fd open.
693 rc = finish_no_open(file, de);
698 ll_intent_release(it);
699 OBD_FREE(it, sizeof(*it));
704 #else /* !HAVE_IOP_ATOMIC_OPEN */
705 static struct lookup_intent *
706 ll_convert_intent(struct open_intent *oit, int lookup_flags)
708 struct lookup_intent *it;
712 return ERR_PTR(-ENOMEM);
714 if (lookup_flags & LOOKUP_OPEN) {
716 if (lookup_flags & LOOKUP_CREATE)
717 it->it_op |= IT_CREAT;
718 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
719 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
720 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
722 it->it_op = IT_GETATTR;
728 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
729 struct nameidata *nd)
734 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
735 struct lookup_intent *it;
737 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
738 it = ll_d2d(dentry)->lld_it;
739 ll_d2d(dentry)->lld_it = NULL;
741 if ((nd->flags & LOOKUP_CREATE) &&
742 !(nd->flags & LOOKUP_OPEN))
745 it = ll_convert_intent(&nd->intent.open, nd->flags);
747 RETURN((struct dentry *)it);
750 de = ll_lookup_it(parent, dentry, it);
753 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
754 if (dentry->d_inode &&
755 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
756 if (S_ISFIFO(dentry->d_inode->i_mode)) {
757 // We cannot call open here as it would
760 (struct ptlrpc_request *)
761 it->d.lustre.it_data);
765 nd->intent.open.file->private_data = it;
766 filp = lookup_instantiate_filp(nd,
772 de = (struct dentry *)filp;
775 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
776 // XXX This can only reliably work on assumption
777 // that there are NO hashed negative dentries.
778 ll_d2d(dentry)->lld_it = it;
779 it = NULL; /* Will be freed in ll_create_nd */
780 /* We absolutely depend on ll_create_nd to be
781 * called to not leak this intent and possible
782 * data attached to it */
787 ll_intent_release(it);
788 OBD_FREE(it, sizeof(*it));
791 de = ll_lookup_it(parent, dentry, NULL);
796 #endif /* HAVE_IOP_ATOMIC_OPEN */
798 /* We depend on "mode" being set with the proper file type/umask by now */
799 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
801 struct inode *inode = NULL;
802 struct ptlrpc_request *request = NULL;
803 struct ll_sb_info *sbi = ll_i2sbi(dir);
807 LASSERT(it && it->d.lustre.it_disposition);
809 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
810 request = it->d.lustre.it_data;
811 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
812 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
814 GOTO(out, inode = ERR_PTR(rc));
816 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
818 /* We asked for a lock on the directory, but were granted a
819 * lock on the inode. Since we finally have an inode pointer,
820 * stuff it in the lock. */
821 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
822 PFID(ll_inode2fid(inode)), inode);
823 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
826 ptlrpc_req_finished(request);
831 * By the time this is called, we already have created the directory cache
832 * entry for the new file, but it is so far negative - it has no inode.
834 * We defer creating the OBD object(s) until open, to keep the intent and
835 * non-intent code paths similar, and also because we do not have the MDS
836 * inode number before calling ll_create_node() (which is needed for LOV),
837 * so we would need to do yet another RPC to the MDS to store the LOV EA
838 * data on the MDS. If needed, we would pass the PACKED lmm as data and
839 * lmm_size in datalen (the MDS still has code which will handle that).
841 * If the create succeeds, we fill in the inode information
842 * with d_instantiate().
844 static int ll_create_it(struct inode *dir, struct dentry *dentry,
845 struct lookup_intent *it)
851 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
852 dentry->d_name.len, dentry->d_name.name,
853 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
855 rc = it_open_error(DISP_OPEN_CREATE, it);
859 inode = ll_create_node(dir, it);
861 RETURN(PTR_ERR(inode));
863 d_instantiate(dentry, inode);
867 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
869 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
873 if (body->mbo_valid & OBD_MD_FLMTIME &&
874 body->mbo_mtime > LTIME_S(inode->i_mtime)) {
875 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
876 "\n", PFID(ll_inode2fid(inode)),
877 LTIME_S(inode->i_mtime), body->mbo_mtime);
878 LTIME_S(inode->i_mtime) = body->mbo_mtime;
881 if (body->mbo_valid & OBD_MD_FLCTIME &&
882 body->mbo_ctime > LTIME_S(inode->i_ctime))
883 LTIME_S(inode->i_ctime) = body->mbo_ctime;
886 static int ll_new_node(struct inode *dir, struct dentry *dchild,
887 const char *tgt, umode_t mode, int rdev, __u32 opc)
889 struct qstr *name = &dchild->d_name;
890 struct ptlrpc_request *request = NULL;
891 struct md_op_data *op_data;
892 struct inode *inode = NULL;
893 struct ll_sb_info *sbi = ll_i2sbi(dir);
898 if (unlikely(tgt != NULL))
899 tgt_len = strlen(tgt) + 1;
901 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
902 name->len, 0, opc, NULL);
904 GOTO(err_exit, err = PTR_ERR(op_data));
906 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
907 from_kuid(&init_user_ns, current_fsuid()),
908 from_kgid(&init_user_ns, current_fsgid()),
909 cfs_curproc_cap_pack(), rdev, &request);
910 ll_finish_md_op_data(op_data);
914 ll_update_times(request, dir);
916 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
920 d_instantiate(dchild, inode);
924 ptlrpc_req_finished(request);
929 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
932 struct qstr *name = &dchild->d_name;
936 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
937 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
940 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
941 mode &= ~current_umask();
943 switch (mode & S_IFMT) {
945 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
951 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
962 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
967 #ifdef HAVE_IOP_ATOMIC_OPEN
969 * Plain create. Intent create is handled in atomic_open.
971 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
972 umode_t mode, bool want_excl)
976 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
977 "flags=%u, excl=%d\n", dentry->d_name.len,
978 dentry->d_name.name, PFID(ll_inode2fid(dir)),
979 dir, mode, want_excl);
981 rc = ll_mknod(dir, dentry, mode, 0);
983 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
985 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
986 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
990 #else /* !HAVE_IOP_ATOMIC_OPEN */
991 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
992 ll_umode_t mode, struct nameidata *nd)
994 struct ll_dentry_data *lld = ll_d2d(dentry);
995 struct lookup_intent *it = NULL;
1002 return ll_mknod(dir, dentry, mode, 0);
1006 /* Was there an error? Propagate it! */
1007 if (it->d.lustre.it_status) {
1008 rc = it->d.lustre.it_status;
1012 rc = ll_create_it(dir, dentry, it);
1013 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1016 nd->intent.open.file->private_data = it;
1017 filp = lookup_instantiate_filp(nd, dentry, NULL);
1023 ll_intent_release(it);
1024 OBD_FREE(it, sizeof(*it));
1027 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1031 #endif /* HAVE_IOP_ATOMIC_OPEN */
1033 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1034 const char *oldpath)
1036 struct qstr *name = &dchild->d_name;
1040 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1041 name->len, name->name, PFID(ll_inode2fid(dir)),
1042 dir, 3000, oldpath);
1044 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1045 LUSTRE_OPC_SYMLINK);
1048 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1053 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1054 struct dentry *new_dentry)
1056 struct inode *src = old_dentry->d_inode;
1057 struct qstr *name = &new_dentry->d_name;
1058 struct ll_sb_info *sbi = ll_i2sbi(dir);
1059 struct ptlrpc_request *request = NULL;
1060 struct md_op_data *op_data;
1064 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1065 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1066 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1068 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1069 0, LUSTRE_OPC_ANY, NULL);
1070 if (IS_ERR(op_data))
1071 RETURN(PTR_ERR(op_data));
1073 err = md_link(sbi->ll_md_exp, op_data, &request);
1074 ll_finish_md_op_data(op_data);
1078 ll_update_times(request, dir);
1079 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1082 ptlrpc_req_finished(request);
1086 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1088 struct qstr *name = &dchild->d_name;
1092 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1093 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1095 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1096 mode &= ~current_umask();
1098 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1100 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1102 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1107 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1109 struct qstr *name = &dchild->d_name;
1110 struct ptlrpc_request *request = NULL;
1111 struct md_op_data *op_data;
1115 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1116 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1118 if (unlikely(d_mountpoint(dchild)))
1121 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1122 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1123 if (IS_ERR(op_data))
1124 RETURN(PTR_ERR(op_data));
1126 if (dchild->d_inode != NULL)
1127 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1129 op_data->op_fid2 = op_data->op_fid3;
1130 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1131 ll_finish_md_op_data(op_data);
1133 ll_update_times(request, dir);
1134 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1137 ptlrpc_req_finished(request);
1144 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1146 struct ptlrpc_request *request = NULL;
1147 struct md_op_data *op_data;
1151 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1152 namelen, name, PFID(ll_inode2fid(dir)), dir);
1154 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1155 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1156 if (IS_ERR(op_data))
1157 RETURN(PTR_ERR(op_data));
1158 op_data->op_cli_flags |= CLI_RM_ENTRY;
1159 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1160 ll_finish_md_op_data(op_data);
1162 ll_update_times(request, dir);
1163 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1166 ptlrpc_req_finished(request);
1170 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1172 struct mdt_body *body;
1173 struct lov_mds_md *eadata;
1174 struct lov_stripe_md *lsm = NULL;
1175 struct obd_trans_info oti = { 0 };
1177 struct obd_capa *oc = NULL;
1181 /* req is swabbed so this is safe */
1182 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1183 if (!(body->mbo_valid & OBD_MD_FLEASIZE))
1186 if (body->mbo_eadatasize == 0) {
1187 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1188 GOTO(out, rc = -EPROTO);
1191 /* The MDS sent back the EA because we unlinked the last reference
1192 * to this file. Use this EA to unlink the objects on the OST.
1193 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1194 * check it is complete and sensible. */
1195 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1196 body->mbo_eadatasize);
1197 LASSERT(eadata != NULL);
1199 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->mbo_eadatasize);
1201 CERROR("obd_unpackmd: %d\n", rc);
1204 LASSERT(rc >= sizeof(*lsm));
1208 GOTO(out_free_memmd, rc = -ENOMEM);
1210 oa->o_oi = lsm->lsm_oi;
1211 oa->o_mode = body->mbo_mode & S_IFMT;
1212 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1214 if (body->mbo_valid & OBD_MD_FLCOOKIE) {
1215 oa->o_valid |= OBD_MD_FLCOOKIE;
1216 oti.oti_logcookies =
1217 req_capsule_server_sized_get(&request->rq_pill,
1219 sizeof(struct llog_cookie) *
1220 lsm->lsm_stripe_count);
1221 if (oti.oti_logcookies == NULL) {
1222 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1223 body->mbo_valid &= ~OBD_MD_FLCOOKIE;
1227 if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
1228 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1230 GOTO(out_free_memmd, rc);
1233 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1234 ll_i2mdexp(dir), oc);
1237 CERROR("obd destroy objid "DOSTID" error %d\n",
1238 POSTID(&lsm->lsm_oi), rc);
1240 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1246 /* ll_unlink() doesn't update the inode with the new link count.
1247 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if
1248 * there is any lock existing. They will recycle dentries and inodes
1249 * based upon locks too. b=20433 */
1250 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1252 struct qstr *name = &dchild->d_name;
1253 struct ptlrpc_request *request = NULL;
1254 struct md_op_data *op_data;
1257 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1258 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1261 * XXX: unlink bind mountpoint maybe call to here,
1262 * just check it as vfs_unlink does.
1264 if (unlikely(d_mountpoint(dchild)))
1267 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1268 LUSTRE_OPC_ANY, NULL);
1269 if (IS_ERR(op_data))
1270 RETURN(PTR_ERR(op_data));
1272 if (dchild->d_inode != NULL)
1273 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1275 op_data->op_fid2 = op_data->op_fid3;
1276 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1277 ll_finish_md_op_data(op_data);
1281 ll_update_times(request, dir);
1282 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1284 rc = ll_objects_destroy(request, dir);
1286 ptlrpc_req_finished(request);
1290 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1291 struct inode *tgt, struct dentry *tgt_dchild)
1293 struct qstr *src_name = &src_dchild->d_name;
1294 struct qstr *tgt_name = &tgt_dchild->d_name;
1295 struct ptlrpc_request *request = NULL;
1296 struct ll_sb_info *sbi = ll_i2sbi(src);
1297 struct md_op_data *op_data;
1300 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1301 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1302 src_name->len, src_name->name,
1303 PFID(ll_inode2fid(src)), src, tgt_name->len,
1304 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1306 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1309 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1310 LUSTRE_OPC_ANY, NULL);
1311 if (IS_ERR(op_data))
1312 RETURN(PTR_ERR(op_data));
1314 if (src_dchild->d_inode != NULL)
1315 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1317 if (tgt_dchild->d_inode != NULL)
1318 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1320 err = md_rename(sbi->ll_md_exp, op_data,
1321 src_name->name, src_name->len,
1322 tgt_name->name, tgt_name->len, &request);
1323 ll_finish_md_op_data(op_data);
1325 ll_update_times(request, src);
1326 ll_update_times(request, tgt);
1327 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1328 err = ll_objects_destroy(request, src);
1331 ptlrpc_req_finished(request);
1334 d_move(src_dchild, tgt_dchild);
1339 const struct inode_operations ll_dir_inode_operations = {
1341 #ifdef HAVE_IOP_ATOMIC_OPEN
1342 .atomic_open = ll_atomic_open,
1344 .lookup = ll_lookup_nd,
1345 .create = ll_create_nd,
1346 /* We need all these non-raw things for NFSD, to not patch it. */
1347 .unlink = ll_unlink,
1350 .symlink = ll_symlink,
1352 .rename = ll_rename,
1353 .setattr = ll_setattr,
1354 .getattr = ll_getattr,
1355 .permission = ll_inode_permission,
1356 .setxattr = ll_setxattr,
1357 .getxattr = ll_getxattr,
1358 .listxattr = ll_listxattr,
1359 .removexattr = ll_removexattr,
1360 #ifdef HAVE_IOP_GET_ACL
1361 .get_acl = ll_get_acl,
1365 const struct inode_operations ll_special_inode_operations = {
1366 .setattr = ll_setattr,
1367 .getattr = ll_getattr,
1368 .permission = ll_inode_permission,
1369 .setxattr = ll_setxattr,
1370 .getxattr = ll_getxattr,
1371 .listxattr = ll_listxattr,
1372 .removexattr = ll_removexattr,
1373 #ifdef HAVE_IOP_GET_ACL
1374 .get_acl = ll_get_acl,