4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 #include <linux/sched.h>
36 #include <linux/quotaops.h>
37 #include <linux/highmem.h>
38 #include <linux/pagemap.h>
39 #include <linux/security.h>
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include "../include/obd_support.h"
44 #include "../include/lustre_fid.h"
45 #include "../include/lustre_dlm.h"
46 #include "../include/lustre_ver.h"
47 #include "llite_internal.h"
49 static int ll_create_it(struct inode *dir, struct dentry *dentry,
50 struct lookup_intent *it);
52 /* called from iget5_locked->find_inode() under inode_hash_lock spinlock */
53 static int ll_test_inode(struct inode *inode, void *opaque)
55 struct ll_inode_info *lli = ll_i2info(inode);
56 struct lustre_md *md = opaque;
58 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
59 CERROR("MDS body missing FID\n");
63 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
69 static int ll_set_inode(struct inode *inode, void *opaque)
71 struct ll_inode_info *lli = ll_i2info(inode);
72 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
74 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
75 CERROR("MDS body missing FID\n");
79 lli->lli_fid = body->mbo_fid1;
80 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
81 CERROR("Can not initialize inode " DFID
82 " without object type: valid = %#llx\n",
83 PFID(&lli->lli_fid), body->mbo_valid);
87 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
88 if (unlikely(inode->i_mode == 0)) {
89 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
99 * Get an inode by inode number(@hash), which is already instantiated by
100 * the intent lookup).
102 struct inode *ll_iget(struct super_block *sb, ino_t hash,
103 struct lustre_md *md)
109 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
111 return ERR_PTR(-ENOMEM);
113 if (inode->i_state & I_NEW) {
114 rc = ll_read_inode2(inode, md);
115 if (!rc && S_ISREG(inode->i_mode) &&
116 !ll_i2info(inode)->lli_clob) {
117 CDEBUG(D_INODE, "%s: apply lsm %p to inode "DFID"\n",
118 ll_get_fsname(sb, NULL, 0), md->lsm,
119 PFID(ll_inode2fid(inode)));
120 rc = cl_file_inode_init(inode, md);
123 make_bad_inode(inode);
124 unlock_new_inode(inode);
128 unlock_new_inode(inode);
130 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
131 rc = ll_update_inode(inode, md);
132 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
133 PFID(&md->body->mbo_fid1), inode, rc);
142 static void ll_invalidate_negative_children(struct inode *dir)
144 struct dentry *dentry, *tmp_subdir;
146 spin_lock(&dir->i_lock);
147 hlist_for_each_entry(dentry, &dir->i_dentry, d_u.d_alias) {
148 spin_lock(&dentry->d_lock);
149 if (!list_empty(&dentry->d_subdirs)) {
150 struct dentry *child;
152 list_for_each_entry_safe(child, tmp_subdir,
155 if (d_really_is_negative(child))
156 d_lustre_invalidate(child, 1);
159 spin_unlock(&dentry->d_lock);
161 spin_unlock(&dir->i_lock);
164 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
166 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
169 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
170 void *data, int flag)
172 struct lustre_handle lockh;
176 case LDLM_CB_BLOCKING:
177 ldlm_lock2handle(lock, &lockh);
178 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
180 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
184 case LDLM_CB_CANCELING: {
185 struct inode *inode = ll_inode_from_resource_lock(lock);
186 __u64 bits = lock->l_policy_data.l_inodebits.bits;
188 /* Inode is set to lock->l_resource->lr_lvb_inode
189 * for mdc - bug 24555
191 LASSERT(!lock->l_ast_data);
196 /* Invalidate all dentries associated with this inode */
197 LASSERT(ldlm_is_canceling(lock));
199 if (!fid_res_name_eq(ll_inode2fid(inode),
200 &lock->l_resource->lr_name)) {
201 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
202 PFID(ll_inode2fid(inode)), inode);
206 if (bits & MDS_INODELOCK_XATTR) {
207 if (S_ISDIR(inode->i_mode))
208 ll_i2info(inode)->lli_def_stripe_offset = -1;
209 ll_xattr_cache_destroy(inode);
210 bits &= ~MDS_INODELOCK_XATTR;
213 /* For OPEN locks we differentiate between lock modes
214 * LCK_CR, LCK_CW, LCK_PR - bug 22891
216 if (bits & MDS_INODELOCK_OPEN)
217 ll_have_md_lock(inode, &bits, lock->l_req_mode);
219 if (bits & MDS_INODELOCK_OPEN) {
222 switch (lock->l_req_mode) {
233 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
237 ll_md_real_close(inode, fmode);
240 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
241 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
242 ll_have_md_lock(inode, &bits, LCK_MINMODE);
244 if (bits & MDS_INODELOCK_LAYOUT) {
245 struct cl_object_conf conf = {
246 .coc_opc = OBJECT_CONF_INVALIDATE,
250 rc = ll_layout_conf(inode, &conf);
252 CDEBUG(D_INODE, "cannot invalidate layout of "
254 PFID(ll_inode2fid(inode)), rc);
257 if (bits & MDS_INODELOCK_UPDATE) {
258 struct ll_inode_info *lli = ll_i2info(inode);
260 spin_lock(&lli->lli_lock);
261 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
262 spin_unlock(&lli->lli_lock);
265 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
266 struct ll_inode_info *lli = ll_i2info(inode);
268 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, pfid = "DFID"\n",
269 PFID(ll_inode2fid(inode)), lli,
270 PFID(&lli->lli_pfid));
272 truncate_inode_pages(inode->i_mapping, 0);
274 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
275 struct inode *master_inode = NULL;
279 * This is slave inode, since all of the child
280 * dentry is connected on the master inode, so
281 * we have to invalidate the negative children
284 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
285 PFID(ll_inode2fid(inode)),
286 PFID(&lli->lli_pfid));
288 hash = cl_fid_build_ino(&lli->lli_pfid,
289 ll_need_32bit_api(ll_i2sbi(inode)));
291 master_inode = ilookup5(inode->i_sb, hash,
292 ll_test_inode_by_fid,
293 (void *)&lli->lli_pfid);
294 if (master_inode && !IS_ERR(master_inode)) {
295 ll_invalidate_negative_children(master_inode);
299 ll_invalidate_negative_children(inode);
303 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
304 inode->i_sb->s_root &&
305 !is_root_inode(inode))
306 ll_invalidate_aliases(inode);
318 __u32 ll_i2suppgid(struct inode *i)
320 if (in_group_p(i->i_gid))
321 return (__u32)from_kgid(&init_user_ns, i->i_gid);
326 /* Pack the required supplementary groups into the supplied groups array.
327 * If we don't need to use the groups from the target inode(s) then we
328 * instead pack one or more groups from the user's supplementary group
329 * array in case it might be useful. Not needed if doing an MDS-side upcall.
331 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
335 suppgids[0] = ll_i2suppgid(i1);
338 suppgids[1] = ll_i2suppgid(i2);
344 * try to reuse three types of dentry:
345 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
346 * by concurrent .revalidate).
347 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
348 * be cleared by others calling d_lustre_revalidate).
349 * 3. DISCONNECTED alias.
351 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
353 struct dentry *alias, *discon_alias, *invalid_alias;
355 if (hlist_empty(&inode->i_dentry))
359 invalid_alias = NULL;
361 spin_lock(&inode->i_lock);
362 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
363 LASSERT(alias != dentry);
365 spin_lock(&alias->d_lock);
366 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
367 S_ISDIR(inode->i_mode))
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 spin_unlock(&inode->i_lock);
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)
402 new = ll_find_alias(inode, de);
412 "Reuse dentry %p inode %p refc %d flags %#x\n",
413 new, d_inode(new), d_count(new), new->d_flags);
421 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
422 de, d_inode(de), d_count(de), de->d_flags);
426 static int ll_lookup_it_finish(struct ptlrpc_request *request,
427 struct lookup_intent *it,
428 struct inode *parent, struct dentry **de)
430 struct inode *inode = NULL;
433 struct dentry *alias;
435 /* NB 1 request reference will be taken away by ll_intent_lock()
438 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
440 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
441 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
445 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
447 /* We used to query real size from OSTs here, but actually
448 * this is not needed. For stat() calls size would be updated
449 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
451 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
452 * Everybody else who needs correct file size would call
453 * ll_glimpse_size or some equivalent themselves anyway.
458 alias = ll_splice_alias(inode, *de);
465 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
466 /* we have lookup look - unhide dentry */
467 if (bits & MDS_INODELOCK_LOOKUP)
468 d_lustre_revalidate(*de);
469 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
470 /* If file created on server, don't depend on parent UPDATE
471 * lock to unhide it. It is left hidden and next lookup can
472 * find it in ll_splice_alias.
474 /* Check that parent has UPDATE lock. */
475 struct lookup_intent parent_it = {
477 .it_lock_handle = 0 };
478 struct lu_fid fid = ll_i2info(parent)->lli_fid;
480 /* If it is striped directory, get the real stripe parent */
481 if (unlikely(ll_i2info(parent)->lli_lsm_md)) {
482 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
483 ll_i2info(parent)->lli_lsm_md,
485 (*de)->d_name.len, &fid);
490 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
492 d_lustre_revalidate(*de);
493 ll_intent_release(&parent_it);
498 if (rc != 0 && it->it_op & IT_OPEN)
499 ll_open_cleanup((*de)->d_sb, request);
504 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
505 struct lookup_intent *it)
507 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
508 struct dentry *save = dentry, *retval;
509 struct ptlrpc_request *req = NULL;
510 struct md_op_data *op_data = NULL;
515 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
516 return ERR_PTR(-ENAMETOOLONG);
518 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p),intent=%s\n",
519 dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
521 if (d_mountpoint(dentry))
522 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
524 if (!it || it->it_op == IT_GETXATTR)
527 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
528 rc = ll_statahead(parent, &dentry, 0);
538 if (it->it_op & IT_CREAT)
539 opc = LUSTRE_OPC_CREATE;
541 opc = LUSTRE_OPC_ANY;
543 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
544 dentry->d_name.len, 0, opc, NULL);
546 return (void *)op_data;
548 /* enforce umask if acl disabled or MDS doesn't support umask */
549 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
550 it->it_create_mode &= ~current_umask();
552 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
553 &ll_md_blocking_ast, 0);
555 * If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
556 * client does not know which suppgid should be sent to the MDS, or
557 * some other(s) changed the target file's GID after this RPC sent
558 * to the MDS with the suppgid as the original GID, then we should
559 * try again with right suppgid.
561 if (rc == -EACCES && it->it_op & IT_OPEN &&
562 it_disposition(it, DISP_OPEN_DENY)) {
563 struct mdt_body *body;
567 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
568 if (op_data->op_suppgids[0] == body->mbo_gid ||
569 op_data->op_suppgids[1] == body->mbo_gid ||
570 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid))) {
571 retval = ERR_PTR(-EACCES);
575 fid_zero(&op_data->op_fid2);
576 op_data->op_suppgids[1] = body->mbo_gid;
577 ptlrpc_req_finished(req);
579 ll_intent_release(it);
580 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
581 ll_md_blocking_ast, 0);
585 retval = ERR_PTR(rc);
589 rc = ll_lookup_it_finish(req, it, parent, &dentry);
591 ll_intent_release(it);
592 retval = ERR_PTR(rc);
596 inode = d_inode(dentry);
597 if ((it->it_op & IT_OPEN) && inode &&
598 !S_ISREG(inode->i_mode) &&
599 !S_ISDIR(inode->i_mode)) {
600 ll_release_openhandle(inode, it);
602 ll_lookup_finish_locks(it, inode);
609 if (op_data && !IS_ERR(op_data))
610 ll_finish_md_op_data(op_data);
612 ptlrpc_req_finished(req);
616 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
619 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
622 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p),flags=%u\n",
623 dentry, PFID(ll_inode2fid(parent)), parent, flags);
625 /* Optimize away (CREATE && !OPEN). Let .create handle the race.
626 * but only if we have write permissions there, otherwise we need
627 * to proceed with lookup. LU-4185
629 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
630 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
633 if (flags & (LOOKUP_PARENT | LOOKUP_OPEN | LOOKUP_CREATE))
637 de = ll_lookup_it(parent, dentry, itp);
640 ll_intent_release(itp);
646 * For cached negative dentry and new dentry, handle lookup/create/open
649 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
650 struct file *file, unsigned open_flags,
651 umode_t mode, int *opened)
653 struct lookup_intent *it;
657 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p),file %p,open_flags %x,mode %x opened %d\n",
658 dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
661 /* Only negative dentries enter here */
662 LASSERT(!d_inode(dentry));
664 if (!d_in_lookup(dentry)) {
665 /* A valid negative dentry that just passed revalidation,
666 * there's little point to try and open it server-side,
667 * even though there's a minuscle chance it might succeed.
668 * Either way it's a valid race to just return -ENOENT here.
670 if (!(open_flags & O_CREAT))
673 /* Otherwise we just unhash it to be rehashed afresh via
674 * lookup if necessary
679 it = kzalloc(sizeof(*it), GFP_NOFS);
684 if (open_flags & O_CREAT)
685 it->it_op |= IT_CREAT;
686 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
687 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
688 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
690 /* Dentry added to dcache tree in ll_lookup_it */
691 de = ll_lookup_it(dir, dentry, it);
698 if (it_disposition(it, DISP_OPEN_CREATE)) {
699 /* Dentry instantiated in ll_create_it. */
700 rc = ll_create_it(dir, dentry, it);
702 /* We dget in ll_splice_alias. */
708 *opened |= FILE_CREATED;
710 if (d_really_is_positive(dentry) && it_disposition(it, DISP_OPEN_OPEN)) {
712 if (S_ISFIFO(d_inode(dentry)->i_mode)) {
713 /* We cannot call open here as it might
714 * deadlock. This case is unreachable in
715 * practice because of OBD_CONNECT_NODEVOH.
717 rc = finish_no_open(file, de);
719 file->private_data = it;
720 rc = finish_open(file, dentry, NULL, opened);
721 /* We dget in ll_splice_alias. finish_open takes
722 * care of dget for fd open.
728 rc = finish_no_open(file, de);
733 ll_intent_release(it);
739 /* We depend on "mode" being set with the proper file type/umask by now */
740 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
742 struct inode *inode = NULL;
743 struct ptlrpc_request *request = NULL;
744 struct ll_sb_info *sbi = ll_i2sbi(dir);
747 LASSERT(it && it->it_disposition);
749 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
750 request = it->it_request;
751 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
752 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
758 LASSERT(hlist_empty(&inode->i_dentry));
760 /* We asked for a lock on the directory, but were granted a
761 * lock on the inode. Since we finally have an inode pointer,
762 * stuff it in the lock.
764 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
765 PFID(ll_inode2fid(dir)), inode);
766 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
768 ptlrpc_req_finished(request);
773 * By the time this is called, we already have created the directory cache
774 * entry for the new file, but it is so far negative - it has no inode.
776 * We defer creating the OBD object(s) until open, to keep the intent and
777 * non-intent code paths similar, and also because we do not have the MDS
778 * inode number before calling ll_create_node() (which is needed for LOV),
779 * so we would need to do yet another RPC to the MDS to store the LOV EA
780 * data on the MDS. If needed, we would pass the PACKED lmm as data and
781 * lmm_size in datalen (the MDS still has code which will handle that).
783 * If the create succeeds, we fill in the inode information
784 * with d_instantiate().
786 static int ll_create_it(struct inode *dir, struct dentry *dentry,
787 struct lookup_intent *it)
792 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
793 dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
795 rc = it_open_error(DISP_OPEN_CREATE, it);
799 inode = ll_create_node(dir, it);
801 return PTR_ERR(inode);
803 d_instantiate(dentry, inode);
807 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
809 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
813 if (body->mbo_valid & OBD_MD_FLMTIME &&
814 body->mbo_mtime > LTIME_S(inode->i_mtime)) {
815 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to %llu\n",
816 PFID(ll_inode2fid(inode)), LTIME_S(inode->i_mtime),
818 LTIME_S(inode->i_mtime) = body->mbo_mtime;
820 if (body->mbo_valid & OBD_MD_FLCTIME &&
821 body->mbo_ctime > LTIME_S(inode->i_ctime))
822 LTIME_S(inode->i_ctime) = body->mbo_ctime;
825 static int ll_new_node(struct inode *dir, struct dentry *dentry,
826 const char *tgt, umode_t mode, int rdev,
829 struct ptlrpc_request *request = NULL;
830 struct md_op_data *op_data;
831 struct inode *inode = NULL;
832 struct ll_sb_info *sbi = ll_i2sbi(dir);
837 tgt_len = strlen(tgt) + 1;
839 op_data = ll_prep_md_op_data(NULL, dir, NULL,
843 if (IS_ERR(op_data)) {
844 err = PTR_ERR(op_data);
848 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
849 from_kuid(&init_user_ns, current_fsuid()),
850 from_kgid(&init_user_ns, current_fsgid()),
851 cfs_curproc_cap_pack(), rdev, &request);
852 ll_finish_md_op_data(op_data);
853 if (err < 0 && err != -EREMOTE)
857 * If the client doesn't know where to create a subdirectory (or
858 * in case of a race that sends the RPC to the wrong MDS), the
859 * MDS will return -EREMOTE and the client will fetch the layout
860 * of the directory, then create the directory on the right MDT.
862 if (unlikely(err == -EREMOTE)) {
863 struct ll_inode_info *lli = ll_i2info(dir);
864 struct lmv_user_md *lum;
867 ptlrpc_req_finished(request);
870 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
873 /* Update stripe_offset and retry */
874 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
875 } else if (err2 == -ENODATA &&
876 lli->lli_def_stripe_offset != -1) {
878 * If there are no default stripe EA on the MDT, but the
879 * client has default stripe, then it probably means
880 * default stripe EA has just been deleted.
882 lli->lli_def_stripe_offset = -1;
887 ptlrpc_req_finished(request);
892 ll_update_times(request, dir);
894 err = ll_prep_inode(&inode, request, dir->i_sb, NULL);
898 d_instantiate(dentry, inode);
901 ptlrpc_req_finished(request);
906 static int ll_mknod(struct inode *dir, struct dentry *dchild,
907 umode_t mode, dev_t rdev)
911 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
912 dchild, PFID(ll_inode2fid(dir)), dir, mode,
913 old_encode_dev(rdev));
915 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
916 mode &= ~current_umask();
918 switch (mode & S_IFMT) {
920 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
926 err = ll_new_node(dir, dchild, NULL, mode,
927 old_encode_dev(rdev),
938 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
944 * Plain create. Intent create is handled in atomic_open.
946 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
947 umode_t mode, bool want_excl)
951 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
952 dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
954 rc = ll_mknod(dir, dentry, mode, 0);
956 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
958 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
959 dentry, d_unhashed(dentry));
964 /* ll_unlink() doesn't update the inode with the new link count.
965 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
966 * is any lock existing. They will recycle dentries and inodes based upon locks
969 static int ll_unlink(struct inode *dir, struct dentry *dchild)
971 struct ptlrpc_request *request = NULL;
972 struct md_op_data *op_data;
975 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
976 dchild, dir->i_ino, dir->i_generation, dir);
978 op_data = ll_prep_md_op_data(NULL, dir, NULL,
981 0, LUSTRE_OPC_ANY, NULL);
983 return PTR_ERR(op_data);
986 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
988 op_data->op_fid2 = op_data->op_fid3;
989 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
990 ll_finish_md_op_data(op_data);
994 ll_update_times(request, dir);
995 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
998 ptlrpc_req_finished(request);
1002 static int ll_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1006 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir"DFID"(%p)\n",
1007 dentry, PFID(ll_inode2fid(dir)), dir);
1009 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1010 mode &= ~current_umask();
1011 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
1013 err = ll_new_node(dir, dentry, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1015 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1020 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1022 struct ptlrpc_request *request = NULL;
1023 struct md_op_data *op_data;
1026 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1027 dchild, PFID(ll_inode2fid(dir)), dir);
1029 op_data = ll_prep_md_op_data(NULL, dir, NULL,
1030 dchild->d_name.name,
1032 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1033 if (IS_ERR(op_data))
1034 return PTR_ERR(op_data);
1036 if (dchild->d_inode)
1037 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1039 op_data->op_fid2 = op_data->op_fid3;
1040 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1041 ll_finish_md_op_data(op_data);
1043 ll_update_times(request, dir);
1044 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1047 ptlrpc_req_finished(request);
1051 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1052 const char *oldname)
1056 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p),target=%.*s\n",
1057 dentry, PFID(ll_inode2fid(dir)), dir, 3000, oldname);
1059 err = ll_new_node(dir, dentry, oldname, S_IFLNK | S_IRWXUGO,
1060 0, LUSTRE_OPC_SYMLINK);
1063 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1068 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1069 struct dentry *new_dentry)
1071 struct inode *src = d_inode(old_dentry);
1072 struct ll_sb_info *sbi = ll_i2sbi(dir);
1073 struct ptlrpc_request *request = NULL;
1074 struct md_op_data *op_data;
1077 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1078 PFID(ll_inode2fid(src)), src, PFID(ll_inode2fid(dir)), dir,
1081 op_data = ll_prep_md_op_data(NULL, src, dir, new_dentry->d_name.name,
1082 new_dentry->d_name.len,
1083 0, LUSTRE_OPC_ANY, NULL);
1084 if (IS_ERR(op_data))
1085 return PTR_ERR(op_data);
1087 err = md_link(sbi->ll_md_exp, op_data, &request);
1088 ll_finish_md_op_data(op_data);
1092 ll_update_times(request, dir);
1093 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1095 ptlrpc_req_finished(request);
1099 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1100 struct inode *tgt, struct dentry *tgt_dchild,
1103 struct ptlrpc_request *request = NULL;
1104 struct ll_sb_info *sbi = ll_i2sbi(src);
1105 struct md_op_data *op_data;
1112 "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1113 src_dchild, PFID(ll_inode2fid(src)), src,
1114 tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1116 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1117 LUSTRE_OPC_ANY, NULL);
1118 if (IS_ERR(op_data))
1119 return PTR_ERR(op_data);
1121 if (src_dchild->d_inode)
1122 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1123 if (tgt_dchild->d_inode)
1124 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1126 err = md_rename(sbi->ll_md_exp, op_data,
1127 src_dchild->d_name.name,
1128 src_dchild->d_name.len,
1129 tgt_dchild->d_name.name,
1130 tgt_dchild->d_name.len, &request);
1131 ll_finish_md_op_data(op_data);
1133 ll_update_times(request, src);
1134 ll_update_times(request, tgt);
1135 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1138 ptlrpc_req_finished(request);
1140 d_move(src_dchild, tgt_dchild);
1144 const struct inode_operations ll_dir_inode_operations = {
1146 .atomic_open = ll_atomic_open,
1147 .lookup = ll_lookup_nd,
1148 .create = ll_create_nd,
1149 /* We need all these non-raw things for NFSD, to not patch it. */
1150 .unlink = ll_unlink,
1153 .symlink = ll_symlink,
1155 .rename = ll_rename,
1156 .setattr = ll_setattr,
1157 .getattr = ll_getattr,
1158 .permission = ll_inode_permission,
1159 .listxattr = ll_listxattr,
1160 .get_acl = ll_get_acl,
1163 const struct inode_operations ll_special_inode_operations = {
1164 .setattr = ll_setattr,
1165 .getattr = ll_getattr,
1166 .permission = ll_inode_permission,
1167 .listxattr = ll_listxattr,
1168 .get_acl = ll_get_acl,