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_lite.h>
50 #include <lustre_dlm.h>
51 #include <lustre_ver.h>
52 #include "llite_internal.h"
54 static int ll_create_it(struct inode *, struct dentry *,
55 int, struct lookup_intent *);
58 * Check if we have something mounted at the named dchild.
59 * In such a case there would always be dentry present.
61 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
66 if (unlikely(dchild)) {
67 mounted = d_mountpoint(dchild);
69 dchild = d_lookup(dparent, name);
71 mounted = d_mountpoint(dchild);
78 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
82 ldlm_lock_decref(lockh, mode);
88 /* called from iget5_locked->find_inode() under inode_lock spinlock */
89 static int ll_test_inode(struct inode *inode, void *opaque)
91 struct ll_inode_info *lli = ll_i2info(inode);
92 struct lustre_md *md = opaque;
94 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
95 CERROR("MDS body missing FID\n");
99 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
105 static int ll_set_inode(struct inode *inode, void *opaque)
107 struct ll_inode_info *lli = ll_i2info(inode);
108 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
110 if (unlikely(!(body->valid & OBD_MD_FLID))) {
111 CERROR("MDS body missing FID\n");
115 lli->lli_fid = body->fid1;
116 if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
117 CERROR("Can not initialize inode "DFID" without object type: "
118 "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
122 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
123 if (unlikely(inode->i_mode == 0)) {
124 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
135 * Get an inode by inode number (already instantiated by the intent lookup).
136 * Returns inode or NULL
138 struct inode *ll_iget(struct super_block *sb, ino_t hash,
139 struct lustre_md *md)
145 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
148 if (inode->i_state & I_NEW) {
151 ll_read_inode2(inode, md);
152 if (S_ISREG(inode->i_mode) &&
153 ll_i2info(inode)->lli_clob == NULL) {
155 "%s: apply lsm %p to inode "DFID".\n",
156 ll_get_fsname(sb, NULL, 0), md->lsm,
157 PFID(ll_inode2fid(inode)));
158 rc = cl_file_inode_init(inode, md);
161 make_bad_inode(inode);
162 unlock_new_inode(inode);
166 unlock_new_inode(inode);
167 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
168 ll_update_inode(inode, md);
169 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
170 inode, PFID(&md->body->fid1));
175 static void ll_invalidate_negative_children(struct inode *dir)
177 struct dentry *dentry, *tmp_subdir;
178 struct ll_d_hlist_node *p;
181 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
182 spin_lock(&dentry->d_lock);
183 if (!list_empty(&dentry->d_subdirs)) {
184 struct dentry *child;
186 list_for_each_entry_safe(child, tmp_subdir,
189 if (child->d_inode == NULL)
190 d_lustre_invalidate(child, 1);
193 spin_unlock(&dentry->d_lock);
195 ll_unlock_dcache(dir);
198 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
199 void *data, int flag)
202 struct lustre_handle lockh;
206 case LDLM_CB_BLOCKING:
207 ldlm_lock2handle(lock, &lockh);
208 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
210 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
214 case LDLM_CB_CANCELING: {
215 struct inode *inode = ll_inode_from_resource_lock(lock);
216 struct ll_inode_info *lli;
217 __u64 bits = lock->l_policy_data.l_inodebits.bits;
219 ldlm_mode_t mode = lock->l_req_mode;
221 /* Inode is set to lock->l_resource->lr_lvb_inode
222 * for mdc - bug 24555 */
223 LASSERT(lock->l_ast_data == NULL);
225 /* Invalidate all dentries associated with this inode */
229 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
230 /* For OPEN locks we differentiate between lock modes
231 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
232 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
233 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
234 ll_have_md_lock(inode, &bits, LCK_MINMODE);
236 if (bits & MDS_INODELOCK_OPEN)
237 ll_have_md_lock(inode, &bits, mode);
239 fid = ll_inode2fid(inode);
240 if (!fid_res_name_eq(fid, &lock->l_resource->lr_name))
241 LDLM_ERROR(lock, "data mismatch with object "
242 DFID" (%p)", PFID(fid), inode);
244 if (bits & MDS_INODELOCK_OPEN) {
246 switch (lock->l_req_mode) {
257 CERROR("Unexpected lock mode for OPEN lock "
258 "%d, inode %ld\n", lock->l_req_mode,
261 ll_md_real_close(inode, flags);
264 lli = ll_i2info(inode);
265 if (bits & MDS_INODELOCK_LAYOUT) {
266 struct cl_object_conf conf = { { 0 } };
268 conf.coc_opc = OBJECT_CONF_INVALIDATE;
269 conf.coc_inode = inode;
270 rc = ll_layout_conf(inode, &conf);
272 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
275 if (bits & MDS_INODELOCK_UPDATE)
276 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
278 if (S_ISDIR(inode->i_mode) &&
279 (bits & MDS_INODELOCK_UPDATE)) {
280 CDEBUG(D_INODE, "invalidating inode %lu\n",
282 truncate_inode_pages(inode->i_mapping, 0);
283 ll_invalidate_negative_children(inode);
286 if (inode->i_sb->s_root &&
287 inode != inode->i_sb->s_root->d_inode &&
288 (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)))
289 ll_invalidate_aliases(inode);
300 __u32 ll_i2suppgid(struct inode *i)
302 if (cfs_curproc_is_in_groups(i->i_gid))
303 return (__u32)i->i_gid;
308 /* Pack the required supplementary groups into the supplied groups array.
309 * If we don't need to use the groups from the target inode(s) then we
310 * instead pack one or more groups from the user's supplementary group
311 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
312 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
319 LASSERT(suppgids != NULL);
321 suppgids[0] = ll_i2suppgid(i1);
324 suppgids[1] = ll_i2suppgid(i2);
329 for (i = 0; i < current_ngroups; i++) {
330 if (suppgids[0] == -1) {
331 if (current_groups[i] != suppgids[1])
332 suppgids[0] = current_groups[i];
335 if (suppgids[1] == -1) {
336 if (current_groups[i] != suppgids[0])
337 suppgids[1] = current_groups[i];
346 * try to reuse three types of dentry:
347 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
348 * by concurrent .revalidate).
349 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
350 * be cleared by others calling d_lustre_revalidate).
351 * 3. DISCONNECTED alias.
353 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
355 struct dentry *alias, *discon_alias, *invalid_alias;
356 struct ll_d_hlist_node *p;
358 if (ll_d_hlist_empty(&inode->i_dentry))
361 discon_alias = invalid_alias = NULL;
363 ll_lock_dcache(inode);
364 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
365 LASSERT(alias != dentry);
367 spin_lock(&alias->d_lock);
368 if (alias->d_flags & DCACHE_DISCONNECTED)
369 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
370 discon_alias = alias;
371 else if (alias->d_parent == dentry->d_parent &&
372 alias->d_name.hash == dentry->d_name.hash &&
373 alias->d_name.len == dentry->d_name.len &&
374 memcmp(alias->d_name.name, dentry->d_name.name,
375 dentry->d_name.len) == 0)
376 invalid_alias = alias;
377 spin_unlock(&alias->d_lock);
382 alias = invalid_alias ?: discon_alias ?: NULL;
384 spin_lock(&alias->d_lock);
386 spin_unlock(&alias->d_lock);
388 ll_unlock_dcache(inode);
394 * Similar to d_splice_alias(), but lustre treats invalid alias
395 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
397 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
402 new = ll_find_alias(inode, de);
404 ll_dops_init(new, 1, 1);
408 "Reuse dentry %p inode %p refc %d flags %#x\n",
409 new, new->d_inode, d_refcount(new), new->d_flags);
413 ll_dops_init(de, 1, 1);
414 __d_lustre_invalidate(de);
416 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
417 de, de->d_inode, d_refcount(de), de->d_flags);
421 int ll_lookup_it_finish(struct ptlrpc_request *request,
422 struct lookup_intent *it, void *data)
424 struct it_cb_data *icbd = data;
425 struct dentry **de = icbd->icbd_childp;
426 struct inode *parent = icbd->icbd_parent;
427 struct inode *inode = NULL;
432 /* NB 1 request reference will be taken away by ll_intent_lock()
434 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
435 it->d.lustre.it_disposition);
436 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
437 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
441 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
443 /* We used to query real size from OSTs here, but actually
444 this is not needed. For stat() calls size would be updated
445 from subsequent do_revalidate()->ll_inode_revalidate_it() in
447 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
448 Everybody else who needs correct file size would call
449 ll_glimpse_size or some equivalent themselves anyway.
450 Also see bug 7198. */
453 /* Only hash *de if it is unhashed (new dentry).
454 * Atoimc_open may passin hashed dentries for open.
457 *de = ll_splice_alias(inode, *de);
459 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
460 /* we have lookup look - unhide dentry */
461 if (bits & MDS_INODELOCK_LOOKUP)
462 d_lustre_revalidate(*de);
463 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
464 /* If file created on server, don't depend on parent UPDATE
465 * lock to unhide it. It is left hidden and next lookup can
466 * find it in ll_splice_alias.
468 /* Check that parent has UPDATE lock. */
469 struct lookup_intent parent_it = {
471 .d.lustre.it_lock_handle = 0 };
473 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
474 &ll_i2info(parent)->lli_fid, NULL)) {
475 d_lustre_revalidate(*de);
476 ll_intent_release(&parent_it);
483 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
484 struct lookup_intent *it, int lookup_flags)
486 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
487 struct dentry *save = dentry, *retval;
488 struct ptlrpc_request *req = NULL;
489 struct md_op_data *op_data;
490 struct it_cb_data icbd;
495 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
496 RETURN(ERR_PTR(-ENAMETOOLONG));
498 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
499 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
500 parent->i_generation, parent, LL_IT2STR(it));
502 if (d_mountpoint(dentry))
503 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
505 ll_frob_intent(&it, &lookup_it);
507 /* As do_lookup is called before follow_mount, root dentry may be left
508 * not valid, revalidate it here. */
509 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
510 (it->it_op & (IT_OPEN | IT_CREAT))) {
511 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
512 MDS_INODELOCK_LOOKUP);
517 if (it->it_op == IT_GETATTR) {
518 rc = ll_statahead_enter(parent, &dentry, 0);
521 GOTO(out, retval = NULL);
522 GOTO(out, retval = dentry);
526 icbd.icbd_childp = &dentry;
527 icbd.icbd_parent = parent;
529 if (it->it_op & IT_CREAT)
530 opc = LUSTRE_OPC_CREATE;
532 opc = LUSTRE_OPC_ANY;
534 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
535 dentry->d_name.len, lookup_flags, opc,
538 RETURN((void *)op_data);
540 /* enforce umask if acl disabled or MDS doesn't support umask */
541 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
542 it->it_create_mode &= ~cfs_curproc_umask();
544 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
545 lookup_flags, &req, ll_md_blocking_ast, 0);
546 ll_finish_md_op_data(op_data);
548 GOTO(out, retval = ERR_PTR(rc));
550 rc = ll_lookup_it_finish(req, it, &icbd);
552 ll_intent_release(it);
553 GOTO(out, retval = ERR_PTR(rc));
556 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
557 !S_ISREG(dentry->d_inode->i_mode) &&
558 !S_ISDIR(dentry->d_inode->i_mode)) {
559 ll_release_openhandle(dentry, it);
561 ll_lookup_finish_locks(it, dentry);
564 GOTO(out, retval = NULL);
566 GOTO(out, retval = dentry);
569 ptlrpc_req_finished(req);
570 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
571 ll_statahead_mark(parent, dentry);
575 #ifdef HAVE_IOP_ATOMIC_OPEN
576 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
579 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
582 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),flags=%u\n",
583 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
584 parent->i_generation, parent, flags);
586 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
587 if ((flags & LOOKUP_CREATE ) && !(flags & LOOKUP_OPEN)) {
588 ll_dops_init(dentry, 1, 1);
589 __d_lustre_invalidate(dentry);
594 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
598 de = ll_lookup_it(parent, dentry, itp, 0);
601 ll_intent_release(itp);
607 * For cached negative dentry and new dentry, handle lookup/create/open
610 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
611 struct file *file, unsigned open_flags,
612 umode_t mode, int *opened)
614 struct lookup_intent *it;
616 long long lookup_flags = LOOKUP_OPEN;
620 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),file %p,"
621 "open_flags %x,mode %x opened %d\n",
622 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
623 dir->i_generation, dir, file, open_flags, mode, *opened);
625 OBD_ALLOC(it, sizeof(*it));
630 if (open_flags & O_CREAT) {
631 it->it_op |= IT_CREAT;
632 lookup_flags |= LOOKUP_CREATE;
634 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
635 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
637 /* Dentry added to dcache tree in ll_lookup_it */
638 de = ll_lookup_it(dir, dentry, it, lookup_flags);
645 if (it_disposition(it, DISP_OPEN_CREATE)) {
646 /* Dentry instantiated in ll_create_it. */
647 rc = ll_create_it(dir, dentry, mode, it);
649 /* We dget in ll_splice_alias. */
655 *opened |= FILE_CREATED;
657 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
659 if (S_ISFIFO(dentry->d_inode->i_mode)) {
660 /* We cannot call open here as it would
663 if (it_disposition(it, DISP_ENQ_OPEN_REF))
665 (struct ptlrpc_request *)
666 it->d.lustre.it_data);
667 rc = finish_no_open(file, de);
669 file->private_data = it;
670 rc = finish_open(file, dentry, NULL, opened);
671 /* We dget in ll_splice_alias. finish_open takes
672 * care of dget for fd open.
678 rc = finish_no_open(file, de);
683 ll_intent_release(it);
684 OBD_FREE(it, sizeof(*it));
689 #else /* !HAVE_IOP_ATOMIC_OPEN */
690 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
693 struct lookup_intent *it;
695 OBD_ALLOC(it, sizeof(*it));
697 return ERR_PTR(-ENOMEM);
699 if (lookup_flags & LOOKUP_OPEN) {
701 if (lookup_flags & LOOKUP_CREATE)
702 it->it_op |= IT_CREAT;
703 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
704 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
706 it->it_op = IT_GETATTR;
712 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
713 struct nameidata *nd)
718 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
719 struct lookup_intent *it;
721 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
722 it = ll_d2d(dentry)->lld_it;
723 ll_d2d(dentry)->lld_it = NULL;
725 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
726 /* We are sure this is new dentry, so we need to create
727 our private data and set the dentry ops */
728 ll_dops_init(dentry, 1, 1);
729 __d_lustre_invalidate(dentry);
733 it = ll_convert_intent(&nd->intent.open, nd->flags);
735 RETURN((struct dentry *)it);
738 de = ll_lookup_it(parent, dentry, it, nd->flags);
741 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
742 if (dentry->d_inode &&
743 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
744 if (S_ISFIFO(dentry->d_inode->i_mode)) {
745 // We cannot call open here as it would
748 (struct ptlrpc_request *)
749 it->d.lustre.it_data);
753 nd->intent.open.file->private_data = it;
754 filp = lookup_instantiate_filp(nd,
760 de = (struct dentry *)filp;
763 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
764 // XXX This can only reliably work on assumption
765 // that there are NO hashed negative dentries.
766 ll_d2d(dentry)->lld_it = it;
767 it = NULL; /* Will be freed in ll_create_nd */
768 /* We absolutely depend on ll_create_nd to be
769 * called to not leak this intent and possible
770 * data attached to it */
775 ll_intent_release(it);
776 OBD_FREE(it, sizeof(*it));
779 de = ll_lookup_it(parent, dentry, NULL, 0);
784 #endif /* HAVE_IOP_ATOMIC_OPEN */
786 /* We depend on "mode" being set with the proper file type/umask by now */
787 static struct inode *ll_create_node(struct inode *dir, const char *name,
788 int namelen, const void *data, int datalen,
789 int mode, __u64 extra,
790 struct lookup_intent *it)
792 struct inode *inode = NULL;
793 struct ptlrpc_request *request = NULL;
794 struct ll_sb_info *sbi = ll_i2sbi(dir);
798 LASSERT(it && it->d.lustre.it_disposition);
800 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
801 request = it->d.lustre.it_data;
802 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
803 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
805 GOTO(out, inode = ERR_PTR(rc));
807 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
809 /* We asked for a lock on the directory, but were granted a
810 * lock on the inode. Since we finally have an inode pointer,
811 * stuff it in the lock. */
812 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
813 inode, inode->i_ino, inode->i_generation);
814 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
817 ptlrpc_req_finished(request);
822 * By the time this is called, we already have created the directory cache
823 * entry for the new file, but it is so far negative - it has no inode.
825 * We defer creating the OBD object(s) until open, to keep the intent and
826 * non-intent code paths similar, and also because we do not have the MDS
827 * inode number before calling ll_create_node() (which is needed for LOV),
828 * so we would need to do yet another RPC to the MDS to store the LOV EA
829 * data on the MDS. If needed, we would pass the PACKED lmm as data and
830 * lmm_size in datalen (the MDS still has code which will handle that).
832 * If the create succeeds, we fill in the inode information
833 * with d_instantiate().
835 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
836 struct lookup_intent *it)
842 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
843 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
844 dir->i_generation, dir, LL_IT2STR(it));
846 rc = it_open_error(DISP_OPEN_CREATE, it);
850 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
851 NULL, 0, mode, 0, it);
853 RETURN(PTR_ERR(inode));
855 if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
856 ll_i2info(inode)->lli_volatile = true;
858 d_instantiate(dentry, inode);
862 static void ll_update_times(struct ptlrpc_request *request,
865 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
869 if (body->valid & OBD_MD_FLMTIME &&
870 body->mtime > LTIME_S(inode->i_mtime)) {
871 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
872 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
873 LTIME_S(inode->i_mtime) = body->mtime;
875 if (body->valid & OBD_MD_FLCTIME &&
876 body->ctime > LTIME_S(inode->i_ctime))
877 LTIME_S(inode->i_ctime) = body->ctime;
880 static int ll_new_node(struct inode *dir, struct qstr *name,
881 const char *tgt, int mode, int rdev,
882 struct dentry *dchild, __u32 opc)
884 struct ptlrpc_request *request = NULL;
885 struct md_op_data *op_data;
886 struct inode *inode = NULL;
887 struct ll_sb_info *sbi = ll_i2sbi(dir);
892 if (unlikely(tgt != NULL))
893 tgt_len = strlen(tgt) + 1;
895 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
896 name->len, 0, opc, NULL);
898 GOTO(err_exit, err = PTR_ERR(op_data));
900 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
901 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
902 cfs_curproc_cap_pack(), rdev, &request);
903 ll_finish_md_op_data(op_data);
907 ll_update_times(request, dir);
910 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
914 d_instantiate(dchild, inode);
918 ptlrpc_req_finished(request);
923 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
924 unsigned rdev, struct dentry *dchild)
929 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
930 name->len, name->name, dir->i_ino, dir->i_generation, dir,
933 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
934 mode &= ~cfs_curproc_umask();
936 switch (mode & S_IFMT) {
938 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
944 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
955 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
960 #ifdef HAVE_IOP_ATOMIC_OPEN
962 * Plain create. Intent create is handled in atomic_open.
964 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
965 umode_t mode, bool want_excl)
969 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),"
970 "flags=%u, excl=%d\n",
971 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
972 dir->i_generation, dir, mode, want_excl);
974 rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
976 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
978 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
979 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
983 #else /* !HAVE_IOP_ATOMIC_OPEN */
984 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
985 ll_umode_t mode, struct nameidata *nd)
987 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
991 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
993 ll_d2d(dentry)->lld_it = NULL;
995 /* Was there an error? Propagate it! */
996 if (it->d.lustre.it_status) {
997 rc = it->d.lustre.it_status;
1001 rc = ll_create_it(dir, dentry, mode, it);
1002 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1005 nd->intent.open.file->private_data = it;
1006 filp = lookup_instantiate_filp(nd, dentry, NULL);
1012 ll_intent_release(it);
1013 OBD_FREE(it, sizeof(*it));
1016 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1020 #endif /* HAVE_IOP_ATOMIC_OPEN */
1022 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1023 const char *tgt, struct dentry *dchild)
1028 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
1029 name->len, name->name, dir->i_ino, dir->i_generation,
1032 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1033 0, dchild, LUSTRE_OPC_SYMLINK);
1036 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1041 static int ll_link_generic(struct inode *src, struct inode *dir,
1042 struct qstr *name, struct dentry *dchild)
1044 struct ll_sb_info *sbi = ll_i2sbi(dir);
1045 struct ptlrpc_request *request = NULL;
1046 struct md_op_data *op_data;
1051 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1052 src->i_ino, src->i_generation, src, dir->i_ino,
1053 dir->i_generation, dir, name->len, name->name);
1055 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1056 0, LUSTRE_OPC_ANY, NULL);
1057 if (IS_ERR(op_data))
1058 RETURN(PTR_ERR(op_data));
1060 err = md_link(sbi->ll_md_exp, op_data, &request);
1061 ll_finish_md_op_data(op_data);
1065 ll_update_times(request, dir);
1066 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1069 ptlrpc_req_finished(request);
1073 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1074 int mode, struct dentry *dchild)
1080 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1081 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1083 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1084 mode &= ~cfs_curproc_umask();
1085 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1086 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1089 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1094 /* Try to find the child dentry by its name.
1095 If found, put the result fid into @fid. */
1096 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1099 struct dentry *parent, *child;
1101 parent = ll_d_hlist_entry(dir->i_dentry, struct dentry, d_alias);
1102 child = d_lookup(parent, name);
1105 *fid = *ll_inode2fid(child->d_inode);
1110 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1111 struct dentry *dchild, struct qstr *name)
1113 struct ptlrpc_request *request = NULL;
1114 struct md_op_data *op_data;
1118 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1119 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1121 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1124 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1125 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1126 if (IS_ERR(op_data))
1127 RETURN(PTR_ERR(op_data));
1129 ll_get_child_fid(dir, name, &op_data->op_fid3);
1130 op_data->op_fid2 = op_data->op_fid3;
1131 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1132 ll_finish_md_op_data(op_data);
1134 ll_update_times(request, dir);
1135 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1138 ptlrpc_req_finished(request);
1145 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1147 struct ptlrpc_request *request = NULL;
1148 struct md_op_data *op_data;
1152 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1153 namelen, name, dir->i_ino, dir->i_generation, dir);
1155 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1156 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1157 if (IS_ERR(op_data))
1158 RETURN(PTR_ERR(op_data));
1159 op_data->op_cli_flags |= CLI_RM_ENTRY;
1160 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1161 ll_finish_md_op_data(op_data);
1163 ll_update_times(request, dir);
1164 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1167 ptlrpc_req_finished(request);
1171 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1173 struct mdt_body *body;
1174 struct lov_mds_md *eadata;
1175 struct lov_stripe_md *lsm = NULL;
1176 struct obd_trans_info oti = { 0 };
1178 struct obd_capa *oc = NULL;
1182 /* req is swabbed so this is safe */
1183 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1184 if (!(body->valid & OBD_MD_FLEASIZE))
1187 if (body->eadatasize == 0) {
1188 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1189 GOTO(out, rc = -EPROTO);
1192 /* The MDS sent back the EA because we unlinked the last reference
1193 * to this file. Use this EA to unlink the objects on the OST.
1194 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1195 * check it is complete and sensible. */
1196 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1198 LASSERT(eadata != NULL);
1200 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1202 CERROR("obd_unpackmd: %d\n", rc);
1205 LASSERT(rc >= sizeof(*lsm));
1209 GOTO(out_free_memmd, rc = -ENOMEM);
1211 oa->o_oi = lsm->lsm_oi;
1212 oa->o_mode = body->mode & S_IFMT;
1213 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1215 if (body->valid & OBD_MD_FLCOOKIE) {
1216 oa->o_valid |= OBD_MD_FLCOOKIE;
1217 oti.oti_logcookies =
1218 req_capsule_server_sized_get(&request->rq_pill,
1220 sizeof(struct llog_cookie) *
1221 lsm->lsm_stripe_count);
1222 if (oti.oti_logcookies == NULL) {
1223 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1224 body->valid &= ~OBD_MD_FLCOOKIE;
1228 if (body->valid & OBD_MD_FLOSSCAPA) {
1229 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1231 GOTO(out_free_memmd, rc);
1234 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1235 ll_i2mdexp(dir), oc);
1238 CERROR("obd destroy objid "DOSTID" error %d\n",
1239 POSTID(&lsm->lsm_oi), rc);
1241 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1247 /* ll_unlink_generic() doesn't update the inode with the new link count.
1248 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1249 * is any lock existing. They will recycle dentries and inodes based upon locks
1251 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1252 struct dentry *dchild, struct qstr *name)
1254 struct ptlrpc_request *request = NULL;
1255 struct md_op_data *op_data;
1258 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1259 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1262 * XXX: unlink bind mountpoint maybe call to here,
1263 * just check it as vfs_unlink does.
1265 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1268 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1269 name->len, 0, LUSTRE_OPC_ANY, NULL);
1270 if (IS_ERR(op_data))
1271 RETURN(PTR_ERR(op_data));
1273 ll_get_child_fid(dir, name, &op_data->op_fid3);
1274 op_data->op_fid2 = op_data->op_fid3;
1275 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1276 ll_finish_md_op_data(op_data);
1280 ll_update_times(request, dir);
1281 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1283 rc = ll_objects_destroy(request, dir);
1285 ptlrpc_req_finished(request);
1289 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1290 struct dentry *src_dchild, struct qstr *src_name,
1291 struct inode *tgt, struct dentry *tgt_dparent,
1292 struct dentry *tgt_dchild, struct qstr *tgt_name)
1294 struct ptlrpc_request *request = NULL;
1295 struct ll_sb_info *sbi = ll_i2sbi(src);
1296 struct md_op_data *op_data;
1299 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1300 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1301 src->i_ino, src->i_generation, src, tgt_name->len,
1302 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1304 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1305 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1308 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1309 LUSTRE_OPC_ANY, NULL);
1310 if (IS_ERR(op_data))
1311 RETURN(PTR_ERR(op_data));
1313 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1314 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1315 err = md_rename(sbi->ll_md_exp, op_data,
1316 src_name->name, src_name->len,
1317 tgt_name->name, tgt_name->len, &request);
1318 ll_finish_md_op_data(op_data);
1320 ll_update_times(request, src);
1321 ll_update_times(request, tgt);
1322 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1323 err = ll_objects_destroy(request, src);
1326 ptlrpc_req_finished(request);
1331 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1334 return ll_mknod_generic(dir, &dchild->d_name, mode,
1335 old_encode_dev(rdev), dchild);
1338 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1340 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1343 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1345 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1348 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1350 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1353 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1354 const char *oldname)
1356 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1359 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1360 struct dentry *new_dentry)
1362 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1366 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1367 struct inode *new_dir, struct dentry *new_dentry)
1370 err = ll_rename_generic(old_dir, NULL,
1371 old_dentry, &old_dentry->d_name,
1372 new_dir, NULL, new_dentry,
1373 &new_dentry->d_name);
1375 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1376 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1378 d_move(old_dentry, new_dentry);
1383 struct inode_operations ll_dir_inode_operations = {
1385 #ifdef HAVE_IOP_ATOMIC_OPEN
1386 .atomic_open = ll_atomic_open,
1388 .lookup = ll_lookup_nd,
1389 .create = ll_create_nd,
1390 /* We need all these non-raw things for NFSD, to not patch it. */
1391 .unlink = ll_unlink,
1394 .symlink = ll_symlink,
1396 .rename = ll_rename,
1397 .setattr = ll_setattr,
1398 .getattr = ll_getattr,
1399 .permission = ll_inode_permission,
1400 .setxattr = ll_setxattr,
1401 .getxattr = ll_getxattr,
1402 .listxattr = ll_listxattr,
1403 .removexattr = ll_removexattr,
1404 #ifdef HAVE_IOP_GET_ACL
1405 .get_acl = ll_get_acl,
1409 struct inode_operations ll_special_inode_operations = {
1410 .setattr = ll_setattr,
1411 .getattr = ll_getattr,
1412 .permission = ll_inode_permission,
1413 .setxattr = ll_setxattr,
1414 .getxattr = ll_getxattr,
1415 .listxattr = ll_listxattr,
1416 .removexattr = ll_removexattr,
1417 #ifdef HAVE_IOP_GET_ACL
1418 .get_acl = ll_get_acl,