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, 2012, 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)
80 ldlm_lock_decref(lockh, mode);
86 /* called from iget5_locked->find_inode() under inode_hash_lock spinlock */
87 static int ll_test_inode(struct inode *inode, void *opaque)
89 struct ll_inode_info *lli = ll_i2info(inode);
90 struct lustre_md *md = opaque;
92 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
93 CERROR("MDS body missing FID\n");
97 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
103 static int ll_set_inode(struct inode *inode, void *opaque)
105 struct ll_inode_info *lli = ll_i2info(inode);
106 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
108 if (unlikely(!(body->valid & OBD_MD_FLID))) {
109 CERROR("MDS body missing FID\n");
113 lli->lli_fid = body->fid1;
114 if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
115 CERROR("Can not initialize inode "DFID" without object type: "
116 "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
120 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
121 if (unlikely(inode->i_mode == 0)) {
122 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
133 * Get an inode by inode number (already instantiated by the intent lookup).
134 * Returns inode or NULL
136 struct inode *ll_iget(struct super_block *sb, ino_t hash,
137 struct lustre_md *md)
142 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
145 if (inode->i_state & I_NEW) {
148 ll_read_inode2(inode, md);
149 if (S_ISREG(inode->i_mode) &&
150 ll_i2info(inode)->lli_clob == NULL) {
152 "%s: apply lsm %p to inode "DFID".\n",
153 ll_get_fsname(sb, NULL, 0), md->lsm,
154 PFID(ll_inode2fid(inode)));
155 rc = cl_file_inode_init(inode, md);
158 make_bad_inode(inode);
159 unlock_new_inode(inode);
163 unlock_new_inode(inode);
164 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
165 ll_update_inode(inode, md);
166 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
167 inode, PFID(&md->body->fid1));
172 static void ll_invalidate_negative_children(struct inode *dir)
174 struct dentry *dentry, *tmp_subdir;
175 struct ll_d_hlist_node *p;
178 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
179 spin_lock(&dentry->d_lock);
180 if (!list_empty(&dentry->d_subdirs)) {
181 struct dentry *child;
183 list_for_each_entry_safe(child, tmp_subdir,
186 if (child->d_inode == NULL)
187 d_lustre_invalidate(child, 1);
190 spin_unlock(&dentry->d_lock);
192 ll_unlock_dcache(dir);
195 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
196 void *data, int flag)
199 struct lustre_handle lockh;
202 case LDLM_CB_BLOCKING:
203 ldlm_lock2handle(lock, &lockh);
204 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
206 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
210 case LDLM_CB_CANCELING: {
211 struct inode *inode = ll_inode_from_resource_lock(lock);
212 struct ll_inode_info *lli;
213 __u64 bits = lock->l_policy_data.l_inodebits.bits;
215 ldlm_mode_t mode = lock->l_req_mode;
217 /* Inode is set to lock->l_resource->lr_lvb_inode
218 * for mdc - bug 24555 */
219 LASSERT(lock->l_ast_data == NULL);
221 /* Invalidate all dentries associated with this inode */
225 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
226 /* For OPEN locks we differentiate between lock modes
227 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
228 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
229 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
230 ll_have_md_lock(inode, &bits, LCK_MINMODE);
232 if (bits & MDS_INODELOCK_OPEN)
233 ll_have_md_lock(inode, &bits, mode);
235 fid = ll_inode2fid(inode);
236 if (!fid_res_name_eq(fid, &lock->l_resource->lr_name))
237 LDLM_ERROR(lock, "data mismatch with object "
238 DFID" (%p)", PFID(fid), inode);
240 if (bits & MDS_INODELOCK_OPEN) {
242 switch (lock->l_req_mode) {
253 CERROR("Unexpected lock mode for OPEN lock "
254 "%d, inode %ld\n", lock->l_req_mode,
257 ll_md_real_close(inode, flags);
260 lli = ll_i2info(inode);
261 if (bits & MDS_INODELOCK_LAYOUT) {
262 struct cl_object_conf conf = { { 0 } };
264 conf.coc_opc = OBJECT_CONF_INVALIDATE;
265 conf.coc_inode = inode;
266 rc = ll_layout_conf(inode, &conf);
268 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
271 if (bits & MDS_INODELOCK_UPDATE) {
272 spin_lock(&lli->lli_lock);
273 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
274 spin_unlock(&lli->lli_lock);
277 if (S_ISDIR(inode->i_mode) &&
278 (bits & MDS_INODELOCK_UPDATE)) {
279 CDEBUG(D_INODE, "invalidating inode %lu\n",
281 truncate_inode_pages(inode->i_mapping, 0);
282 ll_invalidate_negative_children(inode);
285 if (inode->i_sb->s_root &&
286 inode != inode->i_sb->s_root->d_inode &&
287 (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)))
288 ll_invalidate_aliases(inode);
299 __u32 ll_i2suppgid(struct inode *i)
301 if (in_group_p(i->i_gid))
302 return (__u32)from_kgid(&init_user_ns, i->i_gid);
307 /* Pack the required supplementary groups into the supplied groups array.
308 * If we don't need to use the groups from the target inode(s) then we
309 * instead pack one or more groups from the user's supplementary group
310 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
311 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
318 LASSERT(suppgids != NULL);
320 suppgids[0] = ll_i2suppgid(i1);
323 suppgids[1] = ll_i2suppgid(i2);
328 for (i = 0; i < current_ngroups; i++) {
329 if (suppgids[0] == -1) {
330 if (current_groups[i] != suppgids[1])
331 suppgids[0] = current_groups[i];
334 if (suppgids[1] == -1) {
335 if (current_groups[i] != suppgids[0])
336 suppgids[1] = current_groups[i];
345 * try to reuse three types of dentry:
346 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
347 * by concurrent .revalidate).
348 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
349 * be cleared by others calling d_lustre_revalidate).
350 * 3. DISCONNECTED alias.
352 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
354 struct dentry *alias, *discon_alias, *invalid_alias;
355 struct ll_d_hlist_node *p;
357 if (ll_d_hlist_empty(&inode->i_dentry))
360 discon_alias = invalid_alias = NULL;
362 ll_lock_dcache(inode);
363 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
364 LASSERT(alias != dentry);
366 spin_lock(&alias->d_lock);
367 if (alias->d_flags & DCACHE_DISCONNECTED)
368 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
369 discon_alias = alias;
370 else if (alias->d_parent == dentry->d_parent &&
371 alias->d_name.hash == dentry->d_name.hash &&
372 alias->d_name.len == dentry->d_name.len &&
373 memcmp(alias->d_name.name, dentry->d_name.name,
374 dentry->d_name.len) == 0)
375 invalid_alias = alias;
376 spin_unlock(&alias->d_lock);
381 alias = invalid_alias ?: discon_alias ?: NULL;
383 spin_lock(&alias->d_lock);
385 spin_unlock(&alias->d_lock);
387 ll_unlock_dcache(inode);
393 * Similar to d_splice_alias(), but lustre treats invalid alias
394 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
396 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
401 new = ll_find_alias(inode, de);
403 ll_dops_init(new, 1, 1);
407 "Reuse dentry %p inode %p refc %d flags %#x\n",
408 new, new->d_inode, d_count(new), new->d_flags);
412 ll_dops_init(de, 1, 1);
413 __d_lustre_invalidate(de);
415 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
416 de, de->d_inode, d_count(de), de->d_flags);
420 int ll_lookup_it_finish(struct ptlrpc_request *request,
421 struct lookup_intent *it, void *data)
423 struct it_cb_data *icbd = data;
424 struct dentry **de = icbd->icbd_childp;
425 struct inode *parent = icbd->icbd_parent;
426 struct inode *inode = NULL;
430 /* NB 1 request reference will be taken away by ll_intent_lock()
432 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
433 it->d.lustre.it_disposition);
434 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
435 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
439 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
441 /* We used to query real size from OSTs here, but actually
442 this is not needed. For stat() calls size would be updated
443 from subsequent do_revalidate()->ll_inode_revalidate_it() in
445 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
446 Everybody else who needs correct file size would call
447 ll_glimpse_size or some equivalent themselves anyway.
448 Also see bug 7198. */
451 /* Only hash *de if it is unhashed (new dentry).
452 * Atoimc_open may passin hashed dentries for open.
455 *de = ll_splice_alias(inode, *de);
457 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
458 /* we have lookup look - unhide dentry */
459 if (bits & MDS_INODELOCK_LOOKUP)
460 d_lustre_revalidate(*de);
461 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
462 /* If file created on server, don't depend on parent UPDATE
463 * lock to unhide it. It is left hidden and next lookup can
464 * find it in ll_splice_alias.
466 /* Check that parent has UPDATE lock. */
467 struct lookup_intent parent_it = {
469 .d.lustre.it_lock_handle = 0 };
471 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
472 &ll_i2info(parent)->lli_fid, NULL)) {
473 d_lustre_revalidate(*de);
474 ll_intent_release(&parent_it);
481 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
482 struct lookup_intent *it, int lookup_flags)
484 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
485 struct dentry *save = dentry, *retval;
486 struct ptlrpc_request *req = NULL;
487 struct md_op_data *op_data;
488 struct it_cb_data icbd;
492 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
493 return ERR_PTR(-ENAMETOOLONG);
495 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
496 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
497 parent->i_generation, parent, LL_IT2STR(it));
499 if (d_mountpoint(dentry))
500 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
502 ll_frob_intent(&it, &lookup_it);
504 /* As do_lookup is called before follow_mount, root dentry may be left
505 * not valid, revalidate it here. */
506 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
507 (it->it_op & (IT_OPEN | IT_CREAT))) {
508 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
509 MDS_INODELOCK_LOOKUP);
514 if (it->it_op == IT_GETATTR) {
515 rc = ll_statahead_enter(parent, &dentry, 0);
518 GOTO(out, retval = NULL);
519 GOTO(out, retval = dentry);
523 icbd.icbd_childp = &dentry;
524 icbd.icbd_parent = parent;
526 if (it->it_op & IT_CREAT ||
527 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
528 opc = LUSTRE_OPC_CREATE;
530 opc = LUSTRE_OPC_ANY;
532 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
533 dentry->d_name.len, lookup_flags, opc,
536 return (void *)op_data;
538 /* enforce umask if acl disabled or MDS doesn't support umask */
539 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
540 it->it_create_mode &= ~current_umask();
542 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
543 lookup_flags, &req, ll_md_blocking_ast, 0);
544 ll_finish_md_op_data(op_data);
546 GOTO(out, retval = ERR_PTR(rc));
548 rc = ll_lookup_it_finish(req, it, &icbd);
550 ll_intent_release(it);
551 GOTO(out, retval = ERR_PTR(rc));
554 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
555 !S_ISREG(dentry->d_inode->i_mode) &&
556 !S_ISDIR(dentry->d_inode->i_mode)) {
557 ll_release_openhandle(dentry, it);
559 ll_lookup_finish_locks(it, dentry);
562 GOTO(out, retval = NULL);
564 GOTO(out, retval = dentry);
567 ptlrpc_req_finished(req);
568 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
569 ll_statahead_mark(parent, dentry);
573 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
576 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
579 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),flags=%u\n",
580 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
581 parent->i_generation, parent, flags);
583 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
584 if ((flags & LOOKUP_CREATE ) && !(flags & LOOKUP_OPEN)) {
585 ll_dops_init(dentry, 1, 1);
586 __d_lustre_invalidate(dentry);
591 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
595 de = ll_lookup_it(parent, dentry, itp, 0);
598 ll_intent_release(itp);
604 * For cached negative dentry and new dentry, handle lookup/create/open
607 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
608 struct file *file, unsigned open_flags,
609 umode_t mode, int *opened)
611 struct lookup_intent *it;
613 long long lookup_flags = LOOKUP_OPEN;
616 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),file %p,"
617 "open_flags %x,mode %x opened %d\n",
618 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
619 dir->i_generation, dir, file, open_flags, mode, *opened);
621 OBD_ALLOC(it, sizeof(*it));
627 it->it_op |= IT_CREAT;
628 lookup_flags |= LOOKUP_CREATE;
630 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
631 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
633 /* Dentry added to dcache tree in ll_lookup_it */
634 de = ll_lookup_it(dir, dentry, it, lookup_flags);
641 if (it_disposition(it, DISP_OPEN_CREATE)) {
642 /* Dentry instantiated in ll_create_it. */
643 rc = ll_create_it(dir, dentry, mode, it);
645 /* We dget in ll_splice_alias. */
651 *opened |= FILE_CREATED;
653 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
655 if (S_ISFIFO(dentry->d_inode->i_mode)) {
656 /* We cannot call open here as it would
659 if (it_disposition(it, DISP_ENQ_OPEN_REF))
661 (struct ptlrpc_request *)
662 it->d.lustre.it_data);
663 rc = finish_no_open(file, de);
665 file->private_data = it;
666 rc = finish_open(file, dentry, NULL, opened);
667 /* We dget in ll_splice_alias. finish_open takes
668 * care of dget for fd open.
674 rc = finish_no_open(file, de);
679 ll_intent_release(it);
680 OBD_FREE(it, sizeof(*it));
686 /* We depend on "mode" being set with the proper file type/umask by now */
687 static struct inode *ll_create_node(struct inode *dir, const char *name,
688 int namelen, const void *data, int datalen,
689 int mode, __u64 extra,
690 struct lookup_intent *it)
692 struct inode *inode = NULL;
693 struct ptlrpc_request *request = NULL;
694 struct ll_sb_info *sbi = ll_i2sbi(dir);
697 LASSERT(it && it->d.lustre.it_disposition);
699 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
700 request = it->d.lustre.it_data;
701 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
702 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
704 GOTO(out, inode = ERR_PTR(rc));
706 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
708 /* We asked for a lock on the directory, but were granted a
709 * lock on the inode. Since we finally have an inode pointer,
710 * stuff it in the lock. */
711 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
712 inode, inode->i_ino, inode->i_generation);
713 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
715 ptlrpc_req_finished(request);
720 * By the time this is called, we already have created the directory cache
721 * entry for the new file, but it is so far negative - it has no inode.
723 * We defer creating the OBD object(s) until open, to keep the intent and
724 * non-intent code paths similar, and also because we do not have the MDS
725 * inode number before calling ll_create_node() (which is needed for LOV),
726 * so we would need to do yet another RPC to the MDS to store the LOV EA
727 * data on the MDS. If needed, we would pass the PACKED lmm as data and
728 * lmm_size in datalen (the MDS still has code which will handle that).
730 * If the create succeeds, we fill in the inode information
731 * with d_instantiate().
733 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
734 struct lookup_intent *it)
739 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
740 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
741 dir->i_generation, dir, LL_IT2STR(it));
743 rc = it_open_error(DISP_OPEN_CREATE, it);
747 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
748 NULL, 0, mode, 0, it);
750 return PTR_ERR(inode);
752 if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
753 ll_i2info(inode)->lli_volatile = true;
755 d_instantiate(dentry, inode);
759 static void ll_update_times(struct ptlrpc_request *request,
762 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
766 if (body->valid & OBD_MD_FLMTIME &&
767 body->mtime > LTIME_S(inode->i_mtime)) {
768 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
769 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
770 LTIME_S(inode->i_mtime) = body->mtime;
772 if (body->valid & OBD_MD_FLCTIME &&
773 body->ctime > LTIME_S(inode->i_ctime))
774 LTIME_S(inode->i_ctime) = body->ctime;
777 static int ll_new_node(struct inode *dir, struct qstr *name,
778 const char *tgt, int mode, int rdev,
779 struct dentry *dchild, __u32 opc)
781 struct ptlrpc_request *request = NULL;
782 struct md_op_data *op_data;
783 struct inode *inode = NULL;
784 struct ll_sb_info *sbi = ll_i2sbi(dir);
788 if (unlikely(tgt != NULL))
789 tgt_len = strlen(tgt) + 1;
791 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
792 name->len, 0, opc, NULL);
794 GOTO(err_exit, err = PTR_ERR(op_data));
796 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
797 from_kuid(&init_user_ns, current_fsuid()),
798 from_kgid(&init_user_ns, current_fsgid()),
799 cfs_curproc_cap_pack(), rdev, &request);
800 ll_finish_md_op_data(op_data);
804 ll_update_times(request, dir);
807 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
811 d_instantiate(dchild, inode);
814 ptlrpc_req_finished(request);
819 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
820 unsigned rdev, struct dentry *dchild)
824 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
825 name->len, name->name, dir->i_ino, dir->i_generation, dir,
828 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
829 mode &= ~current_umask();
831 switch (mode & S_IFMT) {
833 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
839 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
850 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
856 * Plain create. Intent create is handled in atomic_open.
858 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
859 umode_t mode, bool want_excl)
863 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),"
864 "flags=%u, excl=%d\n",
865 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
866 dir->i_generation, dir, mode, want_excl);
868 rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
870 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
872 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
873 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
878 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
879 const char *tgt, struct dentry *dchild)
883 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
884 name->len, name->name, dir->i_ino, dir->i_generation,
887 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
888 0, dchild, LUSTRE_OPC_SYMLINK);
891 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
896 static int ll_link_generic(struct inode *src, struct inode *dir,
897 struct qstr *name, struct dentry *dchild)
899 struct ll_sb_info *sbi = ll_i2sbi(dir);
900 struct ptlrpc_request *request = NULL;
901 struct md_op_data *op_data;
905 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
906 src->i_ino, src->i_generation, src, dir->i_ino,
907 dir->i_generation, dir, name->len, name->name);
909 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
910 0, LUSTRE_OPC_ANY, NULL);
912 return PTR_ERR(op_data);
914 err = md_link(sbi->ll_md_exp, op_data, &request);
915 ll_finish_md_op_data(op_data);
919 ll_update_times(request, dir);
920 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
922 ptlrpc_req_finished(request);
926 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
927 int mode, struct dentry *dchild)
932 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
933 name->len, name->name, dir->i_ino, dir->i_generation, dir);
935 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
936 mode &= ~current_umask();
937 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
938 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
941 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
946 /* Try to find the child dentry by its name.
947 If found, put the result fid into @fid. */
948 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
951 struct dentry *parent, *child;
953 parent = ll_d_hlist_entry(dir->i_dentry, struct dentry, d_alias);
954 child = d_lookup(parent, name);
957 *fid = *ll_inode2fid(child->d_inode);
962 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
963 struct dentry *dchild, struct qstr *name)
965 struct ptlrpc_request *request = NULL;
966 struct md_op_data *op_data;
969 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
970 name->len, name->name, dir->i_ino, dir->i_generation, dir);
972 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
975 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
976 S_IFDIR, LUSTRE_OPC_ANY, NULL);
978 return PTR_ERR(op_data);
980 ll_get_child_fid(dir, name, &op_data->op_fid3);
981 op_data->op_fid2 = op_data->op_fid3;
982 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
983 ll_finish_md_op_data(op_data);
985 ll_update_times(request, dir);
986 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
989 ptlrpc_req_finished(request);
996 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
998 struct ptlrpc_request *request = NULL;
999 struct md_op_data *op_data;
1002 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1003 namelen, name, dir->i_ino, dir->i_generation, dir);
1005 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1006 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1007 if (IS_ERR(op_data))
1008 return PTR_ERR(op_data);
1009 op_data->op_cli_flags |= CLI_RM_ENTRY;
1010 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1011 ll_finish_md_op_data(op_data);
1013 ll_update_times(request, dir);
1014 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1017 ptlrpc_req_finished(request);
1021 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1023 struct mdt_body *body;
1024 struct lov_mds_md *eadata;
1025 struct lov_stripe_md *lsm = NULL;
1026 struct obd_trans_info oti = { 0 };
1028 struct obd_capa *oc = NULL;
1031 /* req is swabbed so this is safe */
1032 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1033 if (!(body->valid & OBD_MD_FLEASIZE))
1036 if (body->eadatasize == 0) {
1037 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1038 GOTO(out, rc = -EPROTO);
1041 /* The MDS sent back the EA because we unlinked the last reference
1042 * to this file. Use this EA to unlink the objects on the OST.
1043 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1044 * check it is complete and sensible. */
1045 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1047 LASSERT(eadata != NULL);
1049 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1051 CERROR("obd_unpackmd: %d\n", rc);
1054 LASSERT(rc >= sizeof(*lsm));
1058 GOTO(out_free_memmd, rc = -ENOMEM);
1060 oa->o_oi = lsm->lsm_oi;
1061 oa->o_mode = body->mode & S_IFMT;
1062 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1064 if (body->valid & OBD_MD_FLCOOKIE) {
1065 oa->o_valid |= OBD_MD_FLCOOKIE;
1066 oti.oti_logcookies =
1067 req_capsule_server_sized_get(&request->rq_pill,
1069 sizeof(struct llog_cookie) *
1070 lsm->lsm_stripe_count);
1071 if (oti.oti_logcookies == NULL) {
1072 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1073 body->valid &= ~OBD_MD_FLCOOKIE;
1077 if (body->valid & OBD_MD_FLOSSCAPA) {
1078 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1080 GOTO(out_free_memmd, rc);
1083 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1084 ll_i2mdexp(dir), oc);
1087 CERROR("obd destroy objid "DOSTID" error %d\n",
1088 POSTID(&lsm->lsm_oi), rc);
1090 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1096 /* ll_unlink_generic() doesn't update the inode with the new link count.
1097 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1098 * is any lock existing. They will recycle dentries and inodes based upon locks
1100 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1101 struct dentry *dchild, struct qstr *name)
1103 struct ptlrpc_request *request = NULL;
1104 struct md_op_data *op_data;
1106 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1107 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1110 * XXX: unlink bind mountpoint maybe call to here,
1111 * just check it as vfs_unlink does.
1113 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1116 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1117 name->len, 0, LUSTRE_OPC_ANY, NULL);
1118 if (IS_ERR(op_data))
1119 return PTR_ERR(op_data);
1121 ll_get_child_fid(dir, name, &op_data->op_fid3);
1122 op_data->op_fid2 = op_data->op_fid3;
1123 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1124 ll_finish_md_op_data(op_data);
1128 ll_update_times(request, dir);
1129 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1131 rc = ll_objects_destroy(request, dir);
1133 ptlrpc_req_finished(request);
1137 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1138 struct dentry *src_dchild, struct qstr *src_name,
1139 struct inode *tgt, struct dentry *tgt_dparent,
1140 struct dentry *tgt_dchild, struct qstr *tgt_name)
1142 struct ptlrpc_request *request = NULL;
1143 struct ll_sb_info *sbi = ll_i2sbi(src);
1144 struct md_op_data *op_data;
1147 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1148 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1149 src->i_ino, src->i_generation, src, tgt_name->len,
1150 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1152 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1153 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1156 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1157 LUSTRE_OPC_ANY, NULL);
1158 if (IS_ERR(op_data))
1159 return PTR_ERR(op_data);
1161 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1162 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1163 err = md_rename(sbi->ll_md_exp, op_data,
1164 src_name->name, src_name->len,
1165 tgt_name->name, tgt_name->len, &request);
1166 ll_finish_md_op_data(op_data);
1168 ll_update_times(request, src);
1169 ll_update_times(request, tgt);
1170 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1171 err = ll_objects_destroy(request, src);
1174 ptlrpc_req_finished(request);
1179 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1182 return ll_mknod_generic(dir, &dchild->d_name, mode,
1183 old_encode_dev(rdev), dchild);
1186 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1188 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1191 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1193 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1196 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1198 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1201 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1202 const char *oldname)
1204 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1207 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1208 struct dentry *new_dentry)
1210 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1214 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1215 struct inode *new_dir, struct dentry *new_dentry)
1218 err = ll_rename_generic(old_dir, NULL,
1219 old_dentry, &old_dentry->d_name,
1220 new_dir, NULL, new_dentry,
1221 &new_dentry->d_name);
1223 d_move(old_dentry, new_dentry);
1228 struct inode_operations ll_dir_inode_operations = {
1230 .atomic_open = ll_atomic_open,
1231 .lookup = ll_lookup_nd,
1232 .create = ll_create_nd,
1233 /* We need all these non-raw things for NFSD, to not patch it. */
1234 .unlink = ll_unlink,
1237 .symlink = ll_symlink,
1239 .rename = ll_rename,
1240 .setattr = ll_setattr,
1241 .getattr = ll_getattr,
1242 .permission = ll_inode_permission,
1243 .setxattr = ll_setxattr,
1244 .getxattr = ll_getxattr,
1245 .listxattr = ll_listxattr,
1246 .removexattr = ll_removexattr,
1247 .get_acl = ll_get_acl,
1250 struct inode_operations ll_special_inode_operations = {
1251 .setattr = ll_setattr,
1252 .getattr = ll_getattr,
1253 .permission = ll_inode_permission,
1254 .setxattr = ll_setxattr,
1255 .getxattr = ll_getxattr,
1256 .listxattr = ll_listxattr,
1257 .removexattr = ll_removexattr,
1258 .get_acl = ll_get_acl,