4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 #include <linux/sched.h>
40 #include <linux/quotaops.h>
41 #include <linux/highmem.h>
42 #include <linux/pagemap.h>
43 #include <linux/security.h>
44 #include <linux/user_namespace.h>
45 #ifdef HAVE_UIDGID_HEADER
46 # include <linux/uidgid.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include <obd_support.h>
52 #include <lustre_fid.h>
53 #include <lustre_dlm.h>
54 #include <lustre_ver.h>
55 #include "llite_internal.h"
57 static int ll_create_it(struct inode *dir, struct dentry *dentry,
58 struct lookup_intent *it);
60 /* called from iget5_locked->find_inode() under inode_lock spinlock */
61 static int ll_test_inode(struct inode *inode, void *opaque)
63 struct ll_inode_info *lli = ll_i2info(inode);
64 struct lustre_md *md = opaque;
66 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
67 CERROR("MDS body missing FID\n");
71 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
77 static int ll_set_inode(struct inode *inode, void *opaque)
79 struct ll_inode_info *lli = ll_i2info(inode);
80 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
82 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
83 CERROR("MDS body missing FID\n");
87 lli->lli_fid = body->mbo_fid1;
88 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
89 CERROR("Can not initialize inode "DFID" without object type: "
91 PFID(&lli->lli_fid), body->mbo_valid);
95 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
96 if (unlikely(inode->i_mode == 0)) {
97 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
108 * Get an inode by inode number(@hash), which is already instantiated by
109 * the intent lookup).
111 struct inode *ll_iget(struct super_block *sb, ino_t hash,
112 struct lustre_md *md)
120 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
122 RETURN(ERR_PTR(-ENOMEM));
124 if (inode->i_state & I_NEW) {
125 rc = ll_read_inode2(inode, md);
126 if (rc == 0 && S_ISREG(inode->i_mode) &&
127 ll_i2info(inode)->lli_clob == NULL)
128 rc = cl_file_inode_init(inode, md);
131 /* Let's clear directory lsm here, otherwise
132 * make_bad_inode() will reset the inode mode
133 * to regular, then ll_clear_inode will not
134 * be able to clear lsm_md */
135 if (S_ISDIR(inode->i_mode))
136 ll_dir_clear_lsm_md(inode);
137 make_bad_inode(inode);
138 unlock_new_inode(inode);
142 unlock_new_inode(inode);
144 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
145 rc = ll_update_inode(inode, md);
146 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
147 PFID(&md->body->mbo_fid1), inode, rc);
149 if (S_ISDIR(inode->i_mode))
150 ll_dir_clear_lsm_md(inode);
159 static void ll_invalidate_negative_children(struct inode *dir)
161 struct dentry *dentry, *tmp_subdir;
162 DECLARE_LL_D_HLIST_NODE_PTR(p);
165 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
166 spin_lock(&dentry->d_lock);
167 if (!list_empty(&dentry->d_subdirs)) {
168 struct dentry *child;
170 list_for_each_entry_safe(child, tmp_subdir,
173 if (child->d_inode == NULL)
174 d_lustre_invalidate(child, 1);
177 spin_unlock(&dentry->d_lock);
179 ll_unlock_dcache(dir);
182 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
184 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
187 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
188 void *data, int flag)
190 struct lustre_handle lockh;
195 case LDLM_CB_BLOCKING:
196 ldlm_lock2handle(lock, &lockh);
197 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
199 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
203 case LDLM_CB_CANCELING: {
204 struct inode *inode = ll_inode_from_resource_lock(lock);
205 __u64 bits = lock->l_policy_data.l_inodebits.bits;
207 /* Inode is set to lock->l_resource->lr_lvb_inode
208 * for mdc - bug 24555 */
209 LASSERT(lock->l_ast_data == NULL);
214 /* Invalidate all dentries associated with this inode */
215 LASSERT(ldlm_is_canceling(lock));
217 if (!fid_res_name_eq(ll_inode2fid(inode),
218 &lock->l_resource->lr_name)) {
219 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
220 PFID(ll_inode2fid(inode)), inode);
224 if (bits & MDS_INODELOCK_XATTR) {
225 if (S_ISDIR(inode->i_mode))
226 ll_i2info(inode)->lli_def_stripe_offset = -1;
227 ll_xattr_cache_destroy(inode);
228 bits &= ~MDS_INODELOCK_XATTR;
231 /* For OPEN locks we differentiate between lock modes
232 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
233 if (bits & MDS_INODELOCK_OPEN)
234 ll_have_md_lock(inode, &bits, lock->l_req_mode);
236 if (bits & MDS_INODELOCK_OPEN) {
239 switch (lock->l_req_mode) {
250 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
254 ll_md_real_close(inode, fmode);
256 bits &= ~MDS_INODELOCK_OPEN;
259 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
260 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
261 ll_have_md_lock(inode, &bits, LCK_MINMODE);
263 if (bits & MDS_INODELOCK_LAYOUT) {
264 struct cl_object_conf conf = {
265 .coc_opc = OBJECT_CONF_INVALIDATE,
269 rc = ll_layout_conf(inode, &conf);
271 CDEBUG(D_INODE, "cannot invalidate layout of "
273 PFID(ll_inode2fid(inode)), rc);
276 if (bits & MDS_INODELOCK_UPDATE) {
277 struct ll_inode_info *lli = ll_i2info(inode);
279 spin_lock(&lli->lli_lock);
280 LTIME_S(inode->i_mtime) = 0;
281 LTIME_S(inode->i_atime) = 0;
282 LTIME_S(inode->i_ctime) = 0;
283 spin_unlock(&lli->lli_lock);
286 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
287 struct ll_inode_info *lli = ll_i2info(inode);
289 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
290 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
291 lli, PFID(&lli->lli_pfid));
292 truncate_inode_pages(inode->i_mapping, 0);
294 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
295 struct inode *master_inode = NULL;
298 /* This is slave inode, since all of the child
299 * dentry is connected on the master inode, so
300 * we have to invalidate the negative children
302 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
303 PFID(ll_inode2fid(inode)),
304 PFID(&lli->lli_pfid));
306 hash = cl_fid_build_ino(&lli->lli_pfid,
307 ll_need_32bit_api(ll_i2sbi(inode)));
309 /* Do not lookup the inode with ilookup5,
310 * otherwise it will cause dead lock,
312 * 1. Client1 send chmod req to the MDT0, then
313 * on MDT0, it enqueues master and all of its
314 * slaves lock, (mdt_attr_set() ->
315 * mdt_lock_slaves()), after gets master and
316 * stripe0 lock, it will send the enqueue req
317 * (for stripe1) to MDT1, then MDT1 finds the
318 * lock has been granted to client2. Then MDT1
319 * sends blocking ast to client2.
321 * 2. At the same time, client2 tries to unlink
322 * the striped dir (rm -rf striped_dir), and
323 * during lookup, it will hold the master inode
324 * of the striped directory, whose inode state
325 * is NEW, then tries to revalidate all of its
326 * slaves, (ll_prep_inode()->ll_iget()->
327 * ll_read_inode2()-> ll_update_inode().). And
328 * it will be blocked on the server side because
331 * 3. Then the client get the blocking_ast req,
332 * cancel the lock, but being blocked if using
333 * ->ilookup5()), because master inode state is
335 master_inode = ilookup5_nowait(inode->i_sb,
336 hash, ll_test_inode_by_fid,
337 (void *)&lli->lli_pfid);
338 if (master_inode != NULL &&
339 !IS_ERR(master_inode)) {
340 ll_invalidate_negative_children(
345 ll_invalidate_negative_children(inode);
349 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
350 inode->i_sb->s_root != NULL &&
351 inode != inode->i_sb->s_root->d_inode)
352 ll_invalidate_aliases(inode);
364 __u32 ll_i2suppgid(struct inode *i)
366 if (in_group_p(i->i_gid))
367 return (__u32)from_kgid(&init_user_ns, i->i_gid);
369 return (__u32) __kgid_val(INVALID_GID);
372 /* Pack the required supplementary groups into the supplied groups array.
373 * If we don't need to use the groups from the target inode(s) then we
374 * instead pack one or more groups from the user's supplementary group
375 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
376 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
379 LASSERT(suppgids != NULL);
381 suppgids[0] = ll_i2suppgid(i1);
384 suppgids[1] = ll_i2suppgid(i2);
390 * try to reuse three types of dentry:
391 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
392 * by concurrent .revalidate).
393 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
394 * be cleared by others calling d_lustre_revalidate).
395 * 3. DISCONNECTED alias.
397 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
399 struct dentry *alias, *discon_alias, *invalid_alias;
400 DECLARE_LL_D_HLIST_NODE_PTR(p);
402 if (ll_d_hlist_empty(&inode->i_dentry))
405 discon_alias = invalid_alias = NULL;
407 ll_lock_dcache(inode);
408 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
409 LASSERT(alias != dentry);
411 spin_lock(&alias->d_lock);
412 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
413 S_ISDIR(inode->i_mode))
414 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
415 discon_alias = alias;
416 else if (alias->d_parent == dentry->d_parent &&
417 alias->d_name.hash == dentry->d_name.hash &&
418 alias->d_name.len == dentry->d_name.len &&
419 memcmp(alias->d_name.name, dentry->d_name.name,
420 dentry->d_name.len) == 0)
421 invalid_alias = alias;
422 spin_unlock(&alias->d_lock);
427 alias = invalid_alias ?: discon_alias ?: NULL;
429 spin_lock(&alias->d_lock);
431 spin_unlock(&alias->d_lock);
433 ll_unlock_dcache(inode);
439 * Similar to d_splice_alias(), but lustre treats invalid alias
440 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
442 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
448 new = ll_find_alias(inode, de);
458 "Reuse dentry %p inode %p refc %d flags %#x\n",
459 new, new->d_inode, ll_d_count(new), new->d_flags);
467 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
468 de, de->d_inode, ll_d_count(de), de->d_flags);
472 static int ll_lookup_it_finish(struct ptlrpc_request *request,
473 struct lookup_intent *it,
474 struct inode *parent, struct dentry **de)
476 struct inode *inode = NULL;
479 struct dentry *alias;
482 /* NB 1 request reference will be taken away by ll_intent_lock()
484 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
486 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
487 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
491 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
493 /* We used to query real size from OSTs here, but actually
494 this is not needed. For stat() calls size would be updated
495 from subsequent do_revalidate()->ll_inode_revalidate_it() in
497 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
498 Everybody else who needs correct file size would call
499 ll_glimpse_size or some equivalent themselves anyway.
500 Also see bug 7198. */
503 /* Only hash *de if it is unhashed (new dentry).
504 * Atoimc_open may passin hashed dentries for open.
506 alias = ll_splice_alias(inode, *de);
508 GOTO(out, rc = PTR_ERR(alias));
512 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
513 /* we have lookup look - unhide dentry */
514 if (bits & MDS_INODELOCK_LOOKUP)
515 d_lustre_revalidate(*de);
516 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
517 /* If file created on server, don't depend on parent UPDATE
518 * lock to unhide it. It is left hidden and next lookup can
519 * find it in ll_splice_alias.
521 /* Check that parent has UPDATE lock. */
522 struct lookup_intent parent_it = {
524 .it_lock_handle = 0 };
525 struct lu_fid fid = ll_i2info(parent)->lli_fid;
527 /* If it is striped directory, get the real stripe parent */
528 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
529 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
530 ll_i2info(parent)->lli_lsm_md,
532 (*de)->d_name.len, &fid);
537 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
539 d_lustre_revalidate(*de);
540 ll_intent_release(&parent_it);
547 if (rc != 0 && it->it_op & IT_OPEN)
548 ll_open_cleanup((*de)->d_sb, request);
553 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
554 struct lookup_intent *it)
556 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
557 struct dentry *save = dentry, *retval;
558 struct ptlrpc_request *req = NULL;
559 struct md_op_data *op_data = NULL;
564 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
565 RETURN(ERR_PTR(-ENAMETOOLONG));
567 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
568 dentry->d_name.len, dentry->d_name.name,
569 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
571 if (d_mountpoint(dentry))
572 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
574 if (it == NULL || it->it_op == IT_GETXATTR)
577 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
578 rc = ll_statahead(parent, &dentry, 0);
580 RETURN(dentry == save ? NULL : dentry);
583 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
584 dentry->d_sb->s_flags & MS_RDONLY)
585 RETURN(ERR_PTR(-EROFS));
587 if (it->it_op & IT_CREAT)
588 opc = LUSTRE_OPC_CREATE;
590 opc = LUSTRE_OPC_ANY;
592 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
593 dentry->d_name.len, 0, opc, NULL);
595 GOTO(out, retval = ERR_CAST(op_data));
597 /* enforce umask if acl disabled or MDS doesn't support umask */
598 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
599 it->it_create_mode &= ~current_umask();
601 if (it->it_op & IT_CREAT &&
602 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
603 rc = ll_dentry_init_security(dentry, it->it_create_mode,
605 &op_data->op_file_secctx_name,
606 &op_data->op_file_secctx,
607 &op_data->op_file_secctx_size);
609 GOTO(out, retval = ERR_PTR(rc));
612 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
613 &ll_md_blocking_ast, 0);
614 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
615 * client does not know which suppgid should be sent to the MDS, or
616 * some other(s) changed the target file's GID after this RPC sent
617 * to the MDS with the suppgid as the original GID, then we should
618 * try again with right suppgid. */
619 if (rc == -EACCES && it->it_op & IT_OPEN &&
620 it_disposition(it, DISP_OPEN_DENY)) {
621 struct mdt_body *body;
623 LASSERT(req != NULL);
625 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
626 if (op_data->op_suppgids[0] == body->mbo_gid ||
627 op_data->op_suppgids[1] == body->mbo_gid ||
628 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
629 GOTO(out, retval = ERR_PTR(-EACCES));
631 fid_zero(&op_data->op_fid2);
632 op_data->op_suppgids[1] = body->mbo_gid;
633 ptlrpc_req_finished(req);
635 ll_intent_release(it);
636 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
637 &ll_md_blocking_ast, 0);
641 GOTO(out, retval = ERR_PTR(rc));
643 rc = ll_lookup_it_finish(req, it, parent, &dentry);
645 ll_intent_release(it);
646 GOTO(out, retval = ERR_PTR(rc));
649 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
650 !S_ISREG(dentry->d_inode->i_mode) &&
651 !S_ISDIR(dentry->d_inode->i_mode)) {
652 ll_release_openhandle(dentry, it);
654 ll_lookup_finish_locks(it, dentry);
656 GOTO(out, retval = (dentry == save) ? NULL : dentry);
659 if (op_data != NULL && !IS_ERR(op_data))
660 ll_finish_md_op_data(op_data);
662 ptlrpc_req_finished(req);
666 #ifdef HAVE_IOP_ATOMIC_OPEN
667 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
670 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
673 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
674 dentry->d_name.len, dentry->d_name.name,
675 PFID(ll_inode2fid(parent)), parent, flags);
677 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
678 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
681 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
685 de = ll_lookup_it(parent, dentry, itp);
688 ll_intent_release(itp);
694 * For cached negative dentry and new dentry, handle lookup/create/open
697 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
698 struct file *file, unsigned open_flags,
699 umode_t mode, int *opened)
701 struct lookup_intent *it;
703 long long lookup_flags = LOOKUP_OPEN;
707 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
708 "open_flags %x, mode %x opened %d\n",
709 dentry->d_name.len, dentry->d_name.name,
710 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
712 /* Only negative dentries enter here */
713 LASSERT(dentry->d_inode == NULL);
715 if (!d_unhashed(dentry)) {
716 /* A valid negative dentry that just passed revalidation,
717 * there's little point to try and open it server-side,
718 * even though there's a minuscule chance it might succeed.
719 * Either way it's a valid race to just return -ENOENT here.
721 if (!(open_flags & O_CREAT))
724 /* Otherwise we just unhash it to be rehashed afresh via
725 * lookup if necessary
730 OBD_ALLOC(it, sizeof(*it));
735 if (open_flags & O_CREAT) {
736 it->it_op |= IT_CREAT;
737 lookup_flags |= LOOKUP_CREATE;
739 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
740 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
741 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
743 /* Dentry added to dcache tree in ll_lookup_it */
744 de = ll_lookup_it(dir, dentry, it);
750 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
753 if (it_disposition(it, DISP_OPEN_CREATE)) {
754 /* Dentry instantiated in ll_create_it. */
755 rc = ll_create_it(dir, dentry, it);
757 /* We dget in ll_splice_alias. */
763 *opened |= FILE_CREATED;
765 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
767 if (S_ISFIFO(dentry->d_inode->i_mode)) {
768 /* We cannot call open here as it might
769 * deadlock. This case is unreachable in
770 * practice because of OBD_CONNECT_NODEVOH. */
771 rc = finish_no_open(file, de);
773 file->private_data = it;
774 rc = finish_open(file, dentry, NULL, opened);
775 /* We dget in ll_splice_alias. finish_open takes
776 * care of dget for fd open.
782 rc = finish_no_open(file, de);
787 ll_intent_release(it);
788 OBD_FREE(it, sizeof(*it));
793 #else /* !HAVE_IOP_ATOMIC_OPEN */
794 static struct lookup_intent *
795 ll_convert_intent(struct open_intent *oit, int lookup_flags)
797 struct lookup_intent *it;
801 return ERR_PTR(-ENOMEM);
803 if (lookup_flags & LOOKUP_OPEN) {
805 if (lookup_flags & LOOKUP_CREATE)
806 it->it_op |= IT_CREAT;
807 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
808 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
809 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
811 it->it_op = IT_GETATTR;
817 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
818 struct nameidata *nd)
823 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
824 struct lookup_intent *it;
826 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
827 it = ll_d2d(dentry)->lld_it;
828 ll_d2d(dentry)->lld_it = NULL;
830 if ((nd->flags & LOOKUP_CREATE) &&
831 !(nd->flags & LOOKUP_OPEN))
834 it = ll_convert_intent(&nd->intent.open, nd->flags);
836 RETURN((struct dentry *)it);
839 de = ll_lookup_it(parent, dentry, it);
842 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
843 if (dentry->d_inode &&
844 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
845 if (S_ISFIFO(dentry->d_inode->i_mode)) {
846 /* We cannot call open here as it might
847 * deadlock. This case is unreachable in
848 * practice because of
849 * OBD_CONNECT_NODEVOH. */
853 nd->intent.open.file->private_data = it;
854 filp = lookup_instantiate_filp(nd,
860 de = (struct dentry *)filp;
863 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
864 // XXX This can only reliably work on assumption
865 // that there are NO hashed negative dentries.
866 ll_d2d(dentry)->lld_it = it;
867 it = NULL; /* Will be freed in ll_create_nd */
868 /* We absolutely depend on ll_create_nd to be
869 * called to not leak this intent and possible
870 * data attached to it */
875 ll_intent_release(it);
876 OBD_FREE(it, sizeof(*it));
879 de = ll_lookup_it(parent, dentry, NULL);
884 #endif /* HAVE_IOP_ATOMIC_OPEN */
886 /* We depend on "mode" being set with the proper file type/umask by now */
887 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
889 struct inode *inode = NULL;
890 struct ptlrpc_request *request = NULL;
891 struct ll_sb_info *sbi = ll_i2sbi(dir);
895 LASSERT(it && it->it_disposition);
897 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
898 request = it->it_request;
899 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
900 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
902 GOTO(out, inode = ERR_PTR(rc));
904 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
906 /* We asked for a lock on the directory, but were granted a
907 * lock on the inode. Since we finally have an inode pointer,
908 * stuff it in the lock. */
909 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
910 PFID(ll_inode2fid(inode)), inode);
911 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
914 ptlrpc_req_finished(request);
919 * By the time this is called, we already have created the directory cache
920 * entry for the new file, but it is so far negative - it has no inode.
922 * We defer creating the OBD object(s) until open, to keep the intent and
923 * non-intent code paths similar, and also because we do not have the MDS
924 * inode number before calling ll_create_node() (which is needed for LOV),
925 * so we would need to do yet another RPC to the MDS to store the LOV EA
926 * data on the MDS. If needed, we would pass the PACKED lmm as data and
927 * lmm_size in datalen (the MDS still has code which will handle that).
929 * If the create succeeds, we fill in the inode information
930 * with d_instantiate().
932 static int ll_create_it(struct inode *dir, struct dentry *dentry,
933 struct lookup_intent *it)
939 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
940 dentry->d_name.len, dentry->d_name.name,
941 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
943 rc = it_open_error(DISP_OPEN_CREATE, it);
947 inode = ll_create_node(dir, it);
949 RETURN(PTR_ERR(inode));
951 d_instantiate(dentry, inode);
953 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
954 rc = ll_inode_init_security(dentry, inode, dir);
962 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
964 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
968 if (body->mbo_valid & OBD_MD_FLMTIME &&
969 body->mbo_mtime > LTIME_S(inode->i_mtime)) {
970 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to %llu"
971 "\n", PFID(ll_inode2fid(inode)),
972 LTIME_S(inode->i_mtime), body->mbo_mtime);
973 LTIME_S(inode->i_mtime) = body->mbo_mtime;
976 if (body->mbo_valid & OBD_MD_FLCTIME &&
977 body->mbo_ctime > LTIME_S(inode->i_ctime))
978 LTIME_S(inode->i_ctime) = body->mbo_ctime;
981 static int ll_new_node(struct inode *dir, struct dentry *dchild,
982 const char *tgt, umode_t mode, int rdev, __u32 opc)
984 struct qstr *name = &dchild->d_name;
985 struct ptlrpc_request *request = NULL;
986 struct md_op_data *op_data;
987 struct inode *inode = NULL;
988 struct ll_sb_info *sbi = ll_i2sbi(dir);
993 if (unlikely(tgt != NULL))
994 tgt_len = strlen(tgt) + 1;
997 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
998 name->len, 0, opc, NULL);
1000 GOTO(err_exit, err = PTR_ERR(op_data));
1002 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1003 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1004 &op_data->op_file_secctx_name,
1005 &op_data->op_file_secctx,
1006 &op_data->op_file_secctx_size);
1008 GOTO(err_exit, err);
1011 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1012 from_kuid(&init_user_ns, current_fsuid()),
1013 from_kgid(&init_user_ns, current_fsgid()),
1014 cfs_curproc_cap_pack(), rdev, &request);
1015 ll_finish_md_op_data(op_data);
1017 if (err < 0 && err != -EREMOTE)
1018 GOTO(err_exit, err);
1020 /* If the client doesn't know where to create a subdirectory (or
1021 * in case of a race that sends the RPC to the wrong MDS), the
1022 * MDS will return -EREMOTE and the client will fetch the layout
1023 * of the directory, then create the directory on the right MDT. */
1024 if (unlikely(err == -EREMOTE)) {
1025 struct ll_inode_info *lli = ll_i2info(dir);
1026 struct lmv_user_md *lum;
1030 ptlrpc_req_finished(request);
1033 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1034 OBD_MD_DEFAULT_MEA);
1036 /* Update stripe_offset and retry */
1037 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1038 } else if (err2 == -ENODATA &&
1039 lli->lli_def_stripe_offset != -1) {
1040 /* If there are no default stripe EA on the MDT, but the
1041 * client has default stripe, then it probably means
1042 * default stripe EA has just been deleted. */
1043 lli->lli_def_stripe_offset = -1;
1045 GOTO(err_exit, err);
1048 ptlrpc_req_finished(request);
1053 ll_update_times(request, dir);
1055 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1057 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1059 GOTO(err_exit, err);
1061 d_instantiate(dchild, inode);
1063 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1064 err = ll_inode_init_security(dchild, inode, dir);
1066 GOTO(err_exit, err);
1071 if (request != NULL)
1072 ptlrpc_req_finished(request);
1074 if (!IS_ERR_OR_NULL(op_data))
1075 ll_finish_md_op_data(op_data);
1080 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1083 struct qstr *name = &dchild->d_name;
1087 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1088 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1091 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1092 mode &= ~current_umask();
1094 switch (mode & S_IFMT) {
1096 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1102 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1113 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1118 #ifdef HAVE_IOP_ATOMIC_OPEN
1120 * Plain create. Intent create is handled in atomic_open.
1122 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1123 umode_t mode, bool want_excl)
1127 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1129 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1130 "flags=%u, excl=%d\n", dentry->d_name.len,
1131 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1132 dir, mode, want_excl);
1134 rc = ll_mknod(dir, dentry, mode, 0);
1136 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1138 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1139 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1143 #else /* !HAVE_IOP_ATOMIC_OPEN */
1144 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1145 ll_umode_t mode, struct nameidata *nd)
1147 struct ll_dentry_data *lld = ll_d2d(dentry);
1148 struct lookup_intent *it = NULL;
1151 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1157 return ll_mknod(dir, dentry, mode, 0);
1161 /* Was there an error? Propagate it! */
1162 if (it->it_status) {
1167 rc = ll_create_it(dir, dentry, it);
1168 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1171 nd->intent.open.file->private_data = it;
1172 filp = lookup_instantiate_filp(nd, dentry, NULL);
1178 ll_intent_release(it);
1179 OBD_FREE(it, sizeof(*it));
1182 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1186 #endif /* HAVE_IOP_ATOMIC_OPEN */
1188 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1189 const char *oldpath)
1191 struct qstr *name = &dchild->d_name;
1195 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1196 name->len, name->name, PFID(ll_inode2fid(dir)),
1197 dir, 3000, oldpath);
1199 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1200 LUSTRE_OPC_SYMLINK);
1203 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1208 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1209 struct dentry *new_dentry)
1211 struct inode *src = old_dentry->d_inode;
1212 struct qstr *name = &new_dentry->d_name;
1213 struct ll_sb_info *sbi = ll_i2sbi(dir);
1214 struct ptlrpc_request *request = NULL;
1215 struct md_op_data *op_data;
1219 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1220 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1221 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1223 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1224 0, LUSTRE_OPC_ANY, NULL);
1225 if (IS_ERR(op_data))
1226 RETURN(PTR_ERR(op_data));
1228 err = md_link(sbi->ll_md_exp, op_data, &request);
1229 ll_finish_md_op_data(op_data);
1233 ll_update_times(request, dir);
1234 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1237 ptlrpc_req_finished(request);
1241 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1243 struct qstr *name = &dchild->d_name;
1247 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1248 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1250 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1251 mode &= ~current_umask();
1253 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1255 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1257 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1262 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1264 struct qstr *name = &dchild->d_name;
1265 struct ptlrpc_request *request = NULL;
1266 struct md_op_data *op_data;
1270 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1271 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1273 if (unlikely(d_mountpoint(dchild)))
1276 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1277 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1278 if (IS_ERR(op_data))
1279 RETURN(PTR_ERR(op_data));
1281 if (dchild->d_inode != NULL)
1282 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1284 op_data->op_fid2 = op_data->op_fid3;
1285 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1286 ll_finish_md_op_data(op_data);
1288 ll_update_times(request, dir);
1289 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1292 ptlrpc_req_finished(request);
1299 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1301 struct ptlrpc_request *request = NULL;
1302 struct md_op_data *op_data;
1306 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1307 namelen, name, PFID(ll_inode2fid(dir)), dir);
1309 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1310 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1311 if (IS_ERR(op_data))
1312 RETURN(PTR_ERR(op_data));
1313 op_data->op_cli_flags |= CLI_RM_ENTRY;
1314 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1315 ll_finish_md_op_data(op_data);
1317 ll_update_times(request, dir);
1318 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1321 ptlrpc_req_finished(request);
1325 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1327 struct qstr *name = &dchild->d_name;
1328 struct ptlrpc_request *request = NULL;
1329 struct md_op_data *op_data;
1332 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1333 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1336 * XXX: unlink bind mountpoint maybe call to here,
1337 * just check it as vfs_unlink does.
1339 if (unlikely(d_mountpoint(dchild)))
1342 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1343 LUSTRE_OPC_ANY, NULL);
1344 if (IS_ERR(op_data))
1345 RETURN(PTR_ERR(op_data));
1347 if (dchild->d_inode != NULL)
1348 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1350 op_data->op_fid2 = op_data->op_fid3;
1351 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1352 ll_finish_md_op_data(op_data);
1356 ll_update_times(request, dir);
1357 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1360 ptlrpc_req_finished(request);
1364 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1365 struct inode *tgt, struct dentry *tgt_dchild)
1367 struct qstr *src_name = &src_dchild->d_name;
1368 struct qstr *tgt_name = &tgt_dchild->d_name;
1369 struct ptlrpc_request *request = NULL;
1370 struct ll_sb_info *sbi = ll_i2sbi(src);
1371 struct md_op_data *op_data;
1374 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1375 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1376 src_name->len, src_name->name,
1377 PFID(ll_inode2fid(src)), src, tgt_name->len,
1378 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1380 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1383 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1384 LUSTRE_OPC_ANY, NULL);
1385 if (IS_ERR(op_data))
1386 RETURN(PTR_ERR(op_data));
1388 if (src_dchild->d_inode != NULL)
1389 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1391 if (tgt_dchild->d_inode != NULL)
1392 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1394 err = md_rename(sbi->ll_md_exp, op_data,
1395 src_name->name, src_name->len,
1396 tgt_name->name, tgt_name->len, &request);
1397 ll_finish_md_op_data(op_data);
1399 ll_update_times(request, src);
1400 ll_update_times(request, tgt);
1401 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1404 ptlrpc_req_finished(request);
1407 d_move(src_dchild, tgt_dchild);
1412 const struct inode_operations ll_dir_inode_operations = {
1414 #ifdef HAVE_IOP_ATOMIC_OPEN
1415 .atomic_open = ll_atomic_open,
1417 .lookup = ll_lookup_nd,
1418 .create = ll_create_nd,
1419 /* We need all these non-raw things for NFSD, to not patch it. */
1420 .unlink = ll_unlink,
1423 .symlink = ll_symlink,
1425 .rename = ll_rename,
1426 .setattr = ll_setattr,
1427 .getattr = ll_getattr,
1428 .permission = ll_inode_permission,
1429 .setxattr = ll_setxattr,
1430 .getxattr = ll_getxattr,
1431 .listxattr = ll_listxattr,
1432 .removexattr = ll_removexattr,
1433 #ifdef HAVE_IOP_GET_ACL
1434 .get_acl = ll_get_acl,
1438 const struct inode_operations ll_special_inode_operations = {
1439 .setattr = ll_setattr,
1440 .getattr = ll_getattr,
1441 .permission = ll_inode_permission,
1442 .setxattr = ll_setxattr,
1443 .getxattr = ll_getxattr,
1444 .listxattr = ll_listxattr,
1445 .removexattr = ll_removexattr,
1446 #ifdef HAVE_IOP_GET_ACL
1447 .get_acl = ll_get_acl,