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, 2016, 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>
40 #include <linux/user_namespace.h>
41 #ifdef HAVE_UIDGID_HEADER
42 # include <linux/uidgid.h>
45 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <lustre_dlm.h>
50 #include <lustre_ver.h>
51 #include "llite_internal.h"
53 static int ll_create_it(struct inode *dir, struct dentry *dentry,
54 struct lookup_intent *it);
56 /* called from iget5_locked->find_inode() under inode_lock spinlock */
57 static int ll_test_inode(struct inode *inode, void *opaque)
59 struct ll_inode_info *lli = ll_i2info(inode);
60 struct lustre_md *md = opaque;
62 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
63 CERROR("MDS body missing FID\n");
67 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
73 static int ll_set_inode(struct inode *inode, void *opaque)
75 struct ll_inode_info *lli = ll_i2info(inode);
76 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
78 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
79 CERROR("MDS body missing FID\n");
83 lli->lli_fid = body->mbo_fid1;
84 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
85 CERROR("Can not initialize inode "DFID" without object type: "
87 PFID(&lli->lli_fid), body->mbo_valid);
91 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
92 if (unlikely(inode->i_mode == 0)) {
93 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
104 * Get an inode by inode number(@hash), which is already instantiated by
105 * the intent lookup).
107 struct inode *ll_iget(struct super_block *sb, ino_t hash,
108 struct lustre_md *md)
116 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
118 RETURN(ERR_PTR(-ENOMEM));
120 if (inode->i_state & I_NEW) {
121 rc = ll_read_inode2(inode, md);
122 if (rc == 0 && S_ISREG(inode->i_mode) &&
123 ll_i2info(inode)->lli_clob == NULL)
124 rc = cl_file_inode_init(inode, md);
127 /* Let's clear directory lsm here, otherwise
128 * make_bad_inode() will reset the inode mode
129 * to regular, then ll_clear_inode will not
130 * be able to clear lsm_md */
131 if (S_ISDIR(inode->i_mode))
132 ll_dir_clear_lsm_md(inode);
133 make_bad_inode(inode);
134 unlock_new_inode(inode);
138 unlock_new_inode(inode);
140 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
141 rc = ll_update_inode(inode, md);
142 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
143 PFID(&md->body->mbo_fid1), inode, rc);
145 if (S_ISDIR(inode->i_mode))
146 ll_dir_clear_lsm_md(inode);
155 static void ll_invalidate_negative_children(struct inode *dir)
157 struct dentry *dentry, *tmp_subdir;
158 DECLARE_LL_D_HLIST_NODE_PTR(p);
161 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
162 spin_lock(&dentry->d_lock);
163 if (!list_empty(&dentry->d_subdirs)) {
164 struct dentry *child;
166 list_for_each_entry_safe(child, tmp_subdir,
169 if (child->d_inode == NULL)
170 d_lustre_invalidate(child, 1);
173 spin_unlock(&dentry->d_lock);
175 ll_unlock_dcache(dir);
178 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
180 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
183 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
184 void *data, int flag)
186 struct lustre_handle lockh;
191 case LDLM_CB_BLOCKING:
192 ldlm_lock2handle(lock, &lockh);
193 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
195 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
199 case LDLM_CB_CANCELING: {
200 struct inode *inode = ll_inode_from_resource_lock(lock);
201 __u64 bits = lock->l_policy_data.l_inodebits.bits;
203 /* Inode is set to lock->l_resource->lr_lvb_inode
204 * for mdc - bug 24555 */
205 LASSERT(lock->l_ast_data == NULL);
210 /* Invalidate all dentries associated with this inode */
211 LASSERT(ldlm_is_canceling(lock));
213 if (!fid_res_name_eq(ll_inode2fid(inode),
214 &lock->l_resource->lr_name)) {
215 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
216 PFID(ll_inode2fid(inode)), inode);
220 if (bits & MDS_INODELOCK_XATTR) {
221 if (S_ISDIR(inode->i_mode))
222 ll_i2info(inode)->lli_def_stripe_offset = -1;
223 ll_xattr_cache_destroy(inode);
224 bits &= ~MDS_INODELOCK_XATTR;
227 /* For OPEN locks we differentiate between lock modes
228 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
229 if (bits & MDS_INODELOCK_OPEN)
230 ll_have_md_lock(inode, &bits, lock->l_req_mode);
232 if (bits & MDS_INODELOCK_OPEN) {
235 switch (lock->l_req_mode) {
246 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
250 ll_md_real_close(inode, fmode);
252 bits &= ~MDS_INODELOCK_OPEN;
255 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
256 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
257 ll_have_md_lock(inode, &bits, LCK_MINMODE);
259 if (bits & MDS_INODELOCK_LAYOUT) {
260 struct cl_object_conf conf = {
261 .coc_opc = OBJECT_CONF_INVALIDATE,
265 rc = ll_layout_conf(inode, &conf);
267 CDEBUG(D_INODE, "cannot invalidate layout of "
269 PFID(ll_inode2fid(inode)), rc);
272 if (bits & MDS_INODELOCK_UPDATE) {
273 struct ll_inode_info *lli = ll_i2info(inode);
275 spin_lock(&lli->lli_lock);
276 LTIME_S(inode->i_mtime) = 0;
277 LTIME_S(inode->i_atime) = 0;
278 LTIME_S(inode->i_ctime) = 0;
279 spin_unlock(&lli->lli_lock);
282 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
283 struct ll_inode_info *lli = ll_i2info(inode);
285 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
286 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
287 lli, PFID(&lli->lli_pfid));
288 truncate_inode_pages(inode->i_mapping, 0);
290 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
291 struct inode *master_inode = NULL;
294 /* This is slave inode, since all of the child
295 * dentry is connected on the master inode, so
296 * we have to invalidate the negative children
298 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
299 PFID(ll_inode2fid(inode)),
300 PFID(&lli->lli_pfid));
302 hash = cl_fid_build_ino(&lli->lli_pfid,
303 ll_need_32bit_api(ll_i2sbi(inode)));
305 /* Do not lookup the inode with ilookup5,
306 * otherwise it will cause dead lock,
308 * 1. Client1 send chmod req to the MDT0, then
309 * on MDT0, it enqueues master and all of its
310 * slaves lock, (mdt_attr_set() ->
311 * mdt_lock_slaves()), after gets master and
312 * stripe0 lock, it will send the enqueue req
313 * (for stripe1) to MDT1, then MDT1 finds the
314 * lock has been granted to client2. Then MDT1
315 * sends blocking ast to client2.
317 * 2. At the same time, client2 tries to unlink
318 * the striped dir (rm -rf striped_dir), and
319 * during lookup, it will hold the master inode
320 * of the striped directory, whose inode state
321 * is NEW, then tries to revalidate all of its
322 * slaves, (ll_prep_inode()->ll_iget()->
323 * ll_read_inode2()-> ll_update_inode().). And
324 * it will be blocked on the server side because
327 * 3. Then the client get the blocking_ast req,
328 * cancel the lock, but being blocked if using
329 * ->ilookup5()), because master inode state is
331 master_inode = ilookup5_nowait(inode->i_sb,
332 hash, ll_test_inode_by_fid,
333 (void *)&lli->lli_pfid);
335 ll_invalidate_negative_children(
340 ll_invalidate_negative_children(inode);
344 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
345 inode->i_sb->s_root != NULL &&
346 inode != inode->i_sb->s_root->d_inode)
347 ll_invalidate_aliases(inode);
359 __u32 ll_i2suppgid(struct inode *i)
361 if (in_group_p(i->i_gid))
362 return (__u32)from_kgid(&init_user_ns, i->i_gid);
364 return (__u32) __kgid_val(INVALID_GID);
367 /* Pack the required supplementary groups into the supplied groups array.
368 * If we don't need to use the groups from the target inode(s) then we
369 * instead pack one or more groups from the user's supplementary group
370 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
371 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
374 LASSERT(suppgids != NULL);
376 suppgids[0] = ll_i2suppgid(i1);
379 suppgids[1] = ll_i2suppgid(i2);
385 * try to reuse three types of dentry:
386 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
387 * by concurrent .revalidate).
388 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
389 * be cleared by others calling d_lustre_revalidate).
390 * 3. DISCONNECTED alias.
392 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
394 struct dentry *alias, *discon_alias, *invalid_alias;
395 DECLARE_LL_D_HLIST_NODE_PTR(p);
397 if (ll_d_hlist_empty(&inode->i_dentry))
400 discon_alias = invalid_alias = NULL;
402 ll_lock_dcache(inode);
403 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
404 LASSERT(alias != dentry);
406 spin_lock(&alias->d_lock);
407 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
408 S_ISDIR(inode->i_mode))
409 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
410 discon_alias = alias;
411 else if (alias->d_parent == dentry->d_parent &&
412 alias->d_name.hash == dentry->d_name.hash &&
413 alias->d_name.len == dentry->d_name.len &&
414 memcmp(alias->d_name.name, dentry->d_name.name,
415 dentry->d_name.len) == 0)
416 invalid_alias = alias;
417 spin_unlock(&alias->d_lock);
422 alias = invalid_alias ?: discon_alias ?: NULL;
424 spin_lock(&alias->d_lock);
426 spin_unlock(&alias->d_lock);
428 ll_unlock_dcache(inode);
434 * Similar to d_splice_alias(), but lustre treats invalid alias
435 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
437 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
443 new = ll_find_alias(inode, de);
453 "Reuse dentry %p inode %p refc %d flags %#x\n",
454 new, new->d_inode, ll_d_count(new), new->d_flags);
462 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
463 de, de->d_inode, ll_d_count(de), de->d_flags);
467 static int ll_lookup_it_finish(struct ptlrpc_request *request,
468 struct lookup_intent *it,
469 struct inode *parent, struct dentry **de)
471 struct inode *inode = NULL;
474 struct dentry *alias;
477 /* NB 1 request reference will be taken away by ll_intent_lock()
479 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
481 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
482 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
486 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
488 /* We used to query real size from OSTs here, but actually
489 this is not needed. For stat() calls size would be updated
490 from subsequent do_revalidate()->ll_inode_revalidate_it() in
492 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
493 Everybody else who needs correct file size would call
494 ll_glimpse_size or some equivalent themselves anyway.
495 Also see bug 7198. */
498 /* Only hash *de if it is unhashed (new dentry).
499 * Atoimc_open may passin hashed dentries for open.
501 alias = ll_splice_alias(inode, *de);
503 GOTO(out, rc = PTR_ERR(alias));
507 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
508 /* we have lookup look - unhide dentry */
509 if (bits & MDS_INODELOCK_LOOKUP)
510 d_lustre_revalidate(*de);
511 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
512 /* If file created on server, don't depend on parent UPDATE
513 * lock to unhide it. It is left hidden and next lookup can
514 * find it in ll_splice_alias.
516 /* Check that parent has UPDATE lock. */
517 struct lookup_intent parent_it = {
519 .it_lock_handle = 0 };
520 struct lu_fid fid = ll_i2info(parent)->lli_fid;
522 /* If it is striped directory, get the real stripe parent */
523 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
524 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
525 ll_i2info(parent)->lli_lsm_md,
527 (*de)->d_name.len, &fid);
532 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
534 d_lustre_revalidate(*de);
535 ll_intent_release(&parent_it);
542 if (rc != 0 && it->it_op & IT_OPEN)
543 ll_open_cleanup((*de)->d_sb, request);
548 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
549 struct lookup_intent *it)
551 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
552 struct dentry *save = dentry, *retval;
553 struct ptlrpc_request *req = NULL;
554 struct md_op_data *op_data = NULL;
559 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
560 RETURN(ERR_PTR(-ENAMETOOLONG));
562 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
563 dentry->d_name.len, dentry->d_name.name,
564 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
566 if (d_mountpoint(dentry))
567 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
569 if (it == NULL || it->it_op == IT_GETXATTR)
572 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
573 rc = ll_statahead(parent, &dentry, 0);
575 RETURN(dentry == save ? NULL : dentry);
578 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
579 dentry->d_sb->s_flags & MS_RDONLY)
580 RETURN(ERR_PTR(-EROFS));
582 if (it->it_op & IT_CREAT)
583 opc = LUSTRE_OPC_CREATE;
585 opc = LUSTRE_OPC_ANY;
587 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
588 dentry->d_name.len, 0, opc, NULL);
590 GOTO(out, retval = ERR_CAST(op_data));
592 /* enforce umask if acl disabled or MDS doesn't support umask */
593 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
594 it->it_create_mode &= ~current_umask();
596 if (it->it_op & IT_CREAT &&
597 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
598 rc = ll_dentry_init_security(dentry, it->it_create_mode,
600 &op_data->op_file_secctx_name,
601 &op_data->op_file_secctx,
602 &op_data->op_file_secctx_size);
604 GOTO(out, retval = ERR_PTR(rc));
607 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
608 &ll_md_blocking_ast, 0);
609 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
610 * client does not know which suppgid should be sent to the MDS, or
611 * some other(s) changed the target file's GID after this RPC sent
612 * to the MDS with the suppgid as the original GID, then we should
613 * try again with right suppgid. */
614 if (rc == -EACCES && it->it_op & IT_OPEN &&
615 it_disposition(it, DISP_OPEN_DENY)) {
616 struct mdt_body *body;
618 LASSERT(req != NULL);
620 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
621 if (op_data->op_suppgids[0] == body->mbo_gid ||
622 op_data->op_suppgids[1] == body->mbo_gid ||
623 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
624 GOTO(out, retval = ERR_PTR(-EACCES));
626 fid_zero(&op_data->op_fid2);
627 op_data->op_suppgids[1] = body->mbo_gid;
628 ptlrpc_req_finished(req);
630 ll_intent_release(it);
631 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
632 &ll_md_blocking_ast, 0);
636 GOTO(out, retval = ERR_PTR(rc));
638 rc = ll_lookup_it_finish(req, it, parent, &dentry);
640 ll_intent_release(it);
641 GOTO(out, retval = ERR_PTR(rc));
644 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
645 !S_ISREG(dentry->d_inode->i_mode) &&
646 !S_ISDIR(dentry->d_inode->i_mode)) {
647 ll_release_openhandle(dentry, it);
649 ll_lookup_finish_locks(it, dentry);
651 GOTO(out, retval = (dentry == save) ? NULL : dentry);
654 if (op_data != NULL && !IS_ERR(op_data))
655 ll_finish_md_op_data(op_data);
657 ptlrpc_req_finished(req);
661 #ifdef HAVE_IOP_ATOMIC_OPEN
662 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
665 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
668 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
669 dentry->d_name.len, dentry->d_name.name,
670 PFID(ll_inode2fid(parent)), parent, flags);
673 * Optimize away (CREATE && !OPEN). Let .create handle the race.
674 * but only if we have write permissions there, otherwise we need
675 * to proceed with lookup. LU-4185
677 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
678 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
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;
831 * Optimize away (CREATE && !OPEN). Let .create handle
832 * the race. But only if we have write permissions
833 * there, otherwise we need to proceed with lookup.
836 if ((nd->flags & LOOKUP_CREATE) &&
837 !(nd->flags & LOOKUP_OPEN) &&
838 (inode_permission(parent,
839 MAY_WRITE | MAY_EXEC) == 0))
842 it = ll_convert_intent(&nd->intent.open, nd->flags);
844 RETURN((struct dentry *)it);
847 de = ll_lookup_it(parent, dentry, it);
850 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
851 if (dentry->d_inode &&
852 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
853 if (S_ISFIFO(dentry->d_inode->i_mode)) {
854 /* We cannot call open here as it might
855 * deadlock. This case is unreachable in
856 * practice because of
857 * OBD_CONNECT_NODEVOH. */
861 nd->intent.open.file->private_data = it;
862 filp = lookup_instantiate_filp(nd,
868 de = (struct dentry *)filp;
871 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
872 // XXX This can only reliably work on assumption
873 // that there are NO hashed negative dentries.
874 ll_d2d(dentry)->lld_it = it;
875 it = NULL; /* Will be freed in ll_create_nd */
876 /* We absolutely depend on ll_create_nd to be
877 * called to not leak this intent and possible
878 * data attached to it */
883 ll_intent_release(it);
884 OBD_FREE(it, sizeof(*it));
887 de = ll_lookup_it(parent, dentry, NULL);
892 #endif /* HAVE_IOP_ATOMIC_OPEN */
894 /* We depend on "mode" being set with the proper file type/umask by now */
895 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
897 struct inode *inode = NULL;
898 struct ptlrpc_request *request = NULL;
899 struct ll_sb_info *sbi = ll_i2sbi(dir);
903 LASSERT(it && it->it_disposition);
905 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
906 request = it->it_request;
907 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
908 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
910 GOTO(out, inode = ERR_PTR(rc));
912 LASSERT(ll_d_hlist_empty(&inode->i_dentry));
914 /* We asked for a lock on the directory, but were granted a
915 * lock on the inode. Since we finally have an inode pointer,
916 * stuff it in the lock. */
917 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
918 PFID(ll_inode2fid(inode)), inode);
919 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
922 ptlrpc_req_finished(request);
927 * By the time this is called, we already have created the directory cache
928 * entry for the new file, but it is so far negative - it has no inode.
930 * We defer creating the OBD object(s) until open, to keep the intent and
931 * non-intent code paths similar, and also because we do not have the MDS
932 * inode number before calling ll_create_node() (which is needed for LOV),
933 * so we would need to do yet another RPC to the MDS to store the LOV EA
934 * data on the MDS. If needed, we would pass the PACKED lmm as data and
935 * lmm_size in datalen (the MDS still has code which will handle that).
937 * If the create succeeds, we fill in the inode information
938 * with d_instantiate().
940 static int ll_create_it(struct inode *dir, struct dentry *dentry,
941 struct lookup_intent *it)
947 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
948 dentry->d_name.len, dentry->d_name.name,
949 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
951 rc = it_open_error(DISP_OPEN_CREATE, it);
955 inode = ll_create_node(dir, it);
957 RETURN(PTR_ERR(inode));
959 d_instantiate(dentry, inode);
961 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
962 rc = ll_inode_init_security(dentry, inode, dir);
970 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
972 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
976 if (body->mbo_valid & OBD_MD_FLMTIME &&
977 body->mbo_mtime > LTIME_S(inode->i_mtime)) {
978 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to %llu"
979 "\n", PFID(ll_inode2fid(inode)),
980 LTIME_S(inode->i_mtime), body->mbo_mtime);
981 LTIME_S(inode->i_mtime) = body->mbo_mtime;
984 if (body->mbo_valid & OBD_MD_FLCTIME &&
985 body->mbo_ctime > LTIME_S(inode->i_ctime))
986 LTIME_S(inode->i_ctime) = body->mbo_ctime;
989 static int ll_new_node(struct inode *dir, struct dentry *dchild,
990 const char *tgt, umode_t mode, int rdev, __u32 opc)
992 struct qstr *name = &dchild->d_name;
993 struct ptlrpc_request *request = NULL;
994 struct md_op_data *op_data;
995 struct inode *inode = NULL;
996 struct ll_sb_info *sbi = ll_i2sbi(dir);
1001 if (unlikely(tgt != NULL))
1002 tgt_len = strlen(tgt) + 1;
1005 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1006 name->len, 0, opc, NULL);
1007 if (IS_ERR(op_data))
1008 GOTO(err_exit, err = PTR_ERR(op_data));
1010 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1011 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1012 &op_data->op_file_secctx_name,
1013 &op_data->op_file_secctx,
1014 &op_data->op_file_secctx_size);
1016 GOTO(err_exit, err);
1019 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1020 from_kuid(&init_user_ns, current_fsuid()),
1021 from_kgid(&init_user_ns, current_fsgid()),
1022 cfs_curproc_cap_pack(), rdev, &request);
1023 ll_finish_md_op_data(op_data);
1025 if (err < 0 && err != -EREMOTE)
1026 GOTO(err_exit, err);
1028 /* If the client doesn't know where to create a subdirectory (or
1029 * in case of a race that sends the RPC to the wrong MDS), the
1030 * MDS will return -EREMOTE and the client will fetch the layout
1031 * of the directory, then create the directory on the right MDT. */
1032 if (unlikely(err == -EREMOTE)) {
1033 struct ll_inode_info *lli = ll_i2info(dir);
1034 struct lmv_user_md *lum;
1038 ptlrpc_req_finished(request);
1041 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1042 OBD_MD_DEFAULT_MEA);
1044 /* Update stripe_offset and retry */
1045 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1046 } else if (err2 == -ENODATA &&
1047 lli->lli_def_stripe_offset != -1) {
1048 /* If there are no default stripe EA on the MDT, but the
1049 * client has default stripe, then it probably means
1050 * default stripe EA has just been deleted. */
1051 lli->lli_def_stripe_offset = -1;
1053 GOTO(err_exit, err);
1056 ptlrpc_req_finished(request);
1061 ll_update_times(request, dir);
1063 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1065 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1067 GOTO(err_exit, err);
1069 d_instantiate(dchild, inode);
1071 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1072 err = ll_inode_init_security(dchild, inode, dir);
1074 GOTO(err_exit, err);
1079 if (request != NULL)
1080 ptlrpc_req_finished(request);
1082 if (!IS_ERR_OR_NULL(op_data))
1083 ll_finish_md_op_data(op_data);
1088 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1091 struct qstr *name = &dchild->d_name;
1095 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1096 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1099 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1100 mode &= ~current_umask();
1102 switch (mode & S_IFMT) {
1104 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1110 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1121 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1126 #ifdef HAVE_IOP_ATOMIC_OPEN
1128 * Plain create. Intent create is handled in atomic_open.
1130 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1131 umode_t mode, bool want_excl)
1135 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1137 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1138 "flags=%u, excl=%d\n", dentry->d_name.len,
1139 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1140 dir, mode, want_excl);
1142 /* Using mknod(2) to create a regular file is designed to not recognize
1143 * volatile file name, so we use ll_mknod() here. */
1144 rc = ll_mknod(dir, dentry, mode, 0);
1146 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1148 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1149 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1153 #else /* !HAVE_IOP_ATOMIC_OPEN */
1154 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1155 ll_umode_t mode, struct nameidata *nd)
1157 struct ll_dentry_data *lld = ll_d2d(dentry);
1158 struct lookup_intent *it = NULL;
1161 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1167 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1168 * so that volatile file name is recoginized.
1169 * Mknod(2), however, is designed to not recognize volatile
1170 * file name to avoid inode leak under orphan directory until
1172 return ll_new_node(dir, dentry, NULL, mode, 0,
1178 /* Was there an error? Propagate it! */
1179 if (it->it_status) {
1184 rc = ll_create_it(dir, dentry, it);
1185 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1188 nd->intent.open.file->private_data = it;
1189 filp = lookup_instantiate_filp(nd, dentry, NULL);
1195 ll_intent_release(it);
1196 OBD_FREE(it, sizeof(*it));
1199 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1203 #endif /* HAVE_IOP_ATOMIC_OPEN */
1205 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1206 const char *oldpath)
1208 struct qstr *name = &dchild->d_name;
1212 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1213 name->len, name->name, PFID(ll_inode2fid(dir)),
1214 dir, 3000, oldpath);
1216 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1217 LUSTRE_OPC_SYMLINK);
1220 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1225 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1226 struct dentry *new_dentry)
1228 struct inode *src = old_dentry->d_inode;
1229 struct qstr *name = &new_dentry->d_name;
1230 struct ll_sb_info *sbi = ll_i2sbi(dir);
1231 struct ptlrpc_request *request = NULL;
1232 struct md_op_data *op_data;
1236 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1237 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1238 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1240 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1241 0, LUSTRE_OPC_ANY, NULL);
1242 if (IS_ERR(op_data))
1243 RETURN(PTR_ERR(op_data));
1245 err = md_link(sbi->ll_md_exp, op_data, &request);
1246 ll_finish_md_op_data(op_data);
1250 ll_update_times(request, dir);
1251 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1254 ptlrpc_req_finished(request);
1258 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1260 struct qstr *name = &dchild->d_name;
1264 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1265 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1267 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1268 mode &= ~current_umask();
1270 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1272 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1274 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1279 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1281 struct qstr *name = &dchild->d_name;
1282 struct ptlrpc_request *request = NULL;
1283 struct md_op_data *op_data;
1287 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1288 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1290 if (unlikely(d_mountpoint(dchild)))
1293 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1294 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1295 if (IS_ERR(op_data))
1296 RETURN(PTR_ERR(op_data));
1298 if (dchild->d_inode != NULL)
1299 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1301 op_data->op_fid2 = op_data->op_fid3;
1302 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1303 ll_finish_md_op_data(op_data);
1305 ll_update_times(request, dir);
1306 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1309 ptlrpc_req_finished(request);
1316 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1318 struct ptlrpc_request *request = NULL;
1319 struct md_op_data *op_data;
1323 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1324 namelen, name, PFID(ll_inode2fid(dir)), dir);
1326 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1327 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1328 if (IS_ERR(op_data))
1329 RETURN(PTR_ERR(op_data));
1330 op_data->op_cli_flags |= CLI_RM_ENTRY;
1331 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1332 ll_finish_md_op_data(op_data);
1334 ll_update_times(request, dir);
1335 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1338 ptlrpc_req_finished(request);
1342 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1344 struct qstr *name = &dchild->d_name;
1345 struct ptlrpc_request *request = NULL;
1346 struct md_op_data *op_data;
1349 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1350 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1353 * XXX: unlink bind mountpoint maybe call to here,
1354 * just check it as vfs_unlink does.
1356 if (unlikely(d_mountpoint(dchild)))
1359 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1360 LUSTRE_OPC_ANY, NULL);
1361 if (IS_ERR(op_data))
1362 RETURN(PTR_ERR(op_data));
1364 if (dchild->d_inode != NULL)
1365 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1367 op_data->op_fid2 = op_data->op_fid3;
1368 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1369 ll_finish_md_op_data(op_data);
1373 ll_update_times(request, dir);
1374 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1377 ptlrpc_req_finished(request);
1381 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1382 struct inode *tgt, struct dentry *tgt_dchild)
1384 struct qstr *src_name = &src_dchild->d_name;
1385 struct qstr *tgt_name = &tgt_dchild->d_name;
1386 struct ptlrpc_request *request = NULL;
1387 struct ll_sb_info *sbi = ll_i2sbi(src);
1388 struct md_op_data *op_data;
1391 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1392 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1393 src_name->len, src_name->name,
1394 PFID(ll_inode2fid(src)), src, tgt_name->len,
1395 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1397 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1400 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1401 LUSTRE_OPC_ANY, NULL);
1402 if (IS_ERR(op_data))
1403 RETURN(PTR_ERR(op_data));
1405 if (src_dchild->d_inode != NULL)
1406 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1408 if (tgt_dchild->d_inode != NULL)
1409 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1411 err = md_rename(sbi->ll_md_exp, op_data,
1412 src_name->name, src_name->len,
1413 tgt_name->name, tgt_name->len, &request);
1414 ll_finish_md_op_data(op_data);
1416 ll_update_times(request, src);
1417 ll_update_times(request, tgt);
1418 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1421 ptlrpc_req_finished(request);
1424 d_move(src_dchild, tgt_dchild);
1429 const struct inode_operations ll_dir_inode_operations = {
1431 #ifdef HAVE_IOP_ATOMIC_OPEN
1432 .atomic_open = ll_atomic_open,
1434 .lookup = ll_lookup_nd,
1435 .create = ll_create_nd,
1436 /* We need all these non-raw things for NFSD, to not patch it. */
1437 .unlink = ll_unlink,
1440 .symlink = ll_symlink,
1442 .rename = ll_rename,
1443 .setattr = ll_setattr,
1444 .getattr = ll_getattr,
1445 .permission = ll_inode_permission,
1446 .setxattr = ll_setxattr,
1447 .getxattr = ll_getxattr,
1448 .listxattr = ll_listxattr,
1449 .removexattr = ll_removexattr,
1450 #ifdef HAVE_IOP_GET_ACL
1451 .get_acl = ll_get_acl,
1455 const struct inode_operations ll_special_inode_operations = {
1456 .setattr = ll_setattr,
1457 .getattr = ll_getattr,
1458 .permission = ll_inode_permission,
1459 .setxattr = ll_setxattr,
1460 .getxattr = ll_getxattr,
1461 .listxattr = ll_listxattr,
1462 .removexattr = ll_removexattr,
1463 #ifdef HAVE_IOP_GET_ACL
1464 .get_acl = ll_get_acl,