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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdt/mdt_lib.c
34 * Lustre Metadata Target (mdt) request unpacking helper.
36 * Author: Peter Braam <braam@clusterfs.com>
37 * Author: Andreas Dilger <adilger@clusterfs.com>
38 * Author: Phil Schwan <phil@clusterfs.com>
39 * Author: Mike Shaver <shaver@clusterfs.com>
40 * Author: Nikita Danilov <nikita@clusterfs.com>
41 * Author: Huang Hua <huanghua@clusterfs.com>
42 * Author: Fan Yong <fanyong@clusterfs.com>
45 #define DEBUG_SUBSYSTEM S_MDS
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
51 #include "mdt_internal.h"
52 #include <uapi/linux/lnet/nidstr.h>
53 #include <lustre_nodemap.h>
55 typedef enum ucred_init_type {
61 static __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
63 return (__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32);
66 void mdt_exit_ucred(struct mdt_thread_info *info)
68 struct lu_ucred *uc = mdt_ucred(info);
69 struct mdt_device *mdt = info->mti_mdt;
72 if (uc->uc_valid != UCRED_INIT) {
73 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
75 put_group_info(uc->uc_ginfo);
78 if (uc->uc_identity) {
79 mdt_identity_put(mdt->mdt_identity_cache,
81 uc->uc_identity = NULL;
83 uc->uc_valid = UCRED_INIT;
87 static int match_nosquash_list(struct rw_semaphore *sem,
88 struct list_head *nidlist,
94 rc = cfs_match_nid(peernid, nidlist);
99 /* root_squash for inter-MDS operations */
100 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
102 struct lu_ucred *ucred = mdt_ucred(info);
103 struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
106 LASSERT(ucred != NULL);
107 if (!squash->rsi_uid || ucred->uc_fsuid)
110 if (match_nosquash_list(&squash->rsi_sem,
111 &squash->rsi_nosquash_nids,
113 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
114 libcfs_nid2str(peernid));
118 CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
119 libcfs_nid2str(peernid),
120 ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
121 squash->rsi_uid, squash->rsi_gid, 0);
123 ucred->uc_fsuid = squash->rsi_uid;
124 ucred->uc_fsgid = squash->rsi_gid;
126 ucred->uc_suppgids[0] = -1;
127 ucred->uc_suppgids[1] = -1;
132 static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
134 struct ptlrpc_request *req = mdt_info_req(info);
135 const char *jobid = mdt_req_get_jobid(req);
137 /* set jobid if specified. */
139 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
141 uc->uc_jobid[0] = '\0';
144 static void ucred_set_nid(struct mdt_thread_info *info, struct lu_ucred *uc)
146 if (info && info->mti_exp && info->mti_exp->exp_connection)
147 uc->uc_nid = info->mti_exp->exp_connection->c_peer.nid;
149 uc->uc_nid = LNET_NID_ANY;
152 static void ucred_set_audit_enabled(struct mdt_thread_info *info,
155 struct lu_nodemap *nodemap = NULL;
158 if (info && info->mti_exp) {
159 nodemap = nodemap_get_from_exp(info->mti_exp);
160 if (nodemap && !IS_ERR(nodemap)) {
161 audit = nodemap->nmf_enable_audit;
162 nodemap_putref(nodemap);
166 uc->uc_enable_audit = audit;
169 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
170 void *buf, bool drop_fs_cap)
172 struct ptlrpc_request *req = mdt_info_req(info);
173 struct mdt_device *mdt = info->mti_mdt;
174 struct ptlrpc_user_desc *pud = req->rq_user_desc;
175 struct lu_ucred *ucred = mdt_ucred(info);
176 struct lu_nodemap *nodemap;
177 lnet_nid_t peernid = req->rq_peer.nid;
185 LASSERT(req->rq_auth_gss);
186 LASSERT(!req->rq_auth_usr_mdt);
187 LASSERT(req->rq_user_desc);
188 LASSERT(ucred != NULL);
190 ucred->uc_valid = UCRED_INVALID;
192 nodemap = nodemap_get_from_exp(info->mti_exp);
194 RETURN(PTR_ERR(nodemap));
196 pud->pud_uid = nodemap_map_id(nodemap, NODEMAP_UID,
197 NODEMAP_CLIENT_TO_FS, pud->pud_uid);
198 pud->pud_gid = nodemap_map_id(nodemap, NODEMAP_GID,
199 NODEMAP_CLIENT_TO_FS, pud->pud_gid);
200 pud->pud_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
201 NODEMAP_CLIENT_TO_FS, pud->pud_fsuid);
202 pud->pud_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
203 NODEMAP_CLIENT_TO_FS, pud->pud_fsgid);
205 ucred->uc_o_uid = pud->pud_uid;
206 ucred->uc_o_gid = pud->pud_gid;
207 ucred->uc_o_fsuid = pud->pud_fsuid;
208 ucred->uc_o_fsgid = pud->pud_fsgid;
210 if (nodemap && ucred->uc_o_uid == nodemap->nm_squash_uid) {
211 /* deny access before we get identity ref */
212 if (nodemap->nmf_deny_unknown) {
213 nodemap_putref(nodemap);
217 ucred->uc_fsuid = nodemap->nm_squash_uid;
218 ucred->uc_fsgid = nodemap->nm_squash_gid;
220 ucred->uc_suppgids[0] = -1;
221 ucred->uc_suppgids[1] = -1;
223 nodemap_putref(nodemap);
225 if (type == BODY_INIT) {
226 struct mdt_body *body = (struct mdt_body *)buf;
228 ucred->uc_suppgids[0] = body->mbo_suppgid;
229 ucred->uc_suppgids[1] = -1;
232 if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
233 req->rq_auth_uid != pud->pud_uid) {
234 CDEBUG(D_SEC, "local client %s: auth uid %u "
235 "while client claims %u:%u/%u:%u\n",
236 libcfs_nid2str(peernid), req->rq_auth_uid,
237 pud->pud_uid, pud->pud_gid,
238 pud->pud_fsuid, pud->pud_fsgid);
242 if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
243 ucred->uc_identity = NULL;
244 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
246 struct md_identity *identity;
248 identity = mdt_identity_get(mdt->mdt_identity_cache,
250 if (IS_ERR(identity)) {
251 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
252 ucred->uc_identity = NULL;
253 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
257 "Deny access without identity: uid %u\n",
262 ucred->uc_identity = identity;
263 perm = mdt_identity_get_perm(ucred->uc_identity,
268 /* find out the setuid/setgid attempt */
269 setuid = (pud->pud_uid != pud->pud_fsuid);
270 setgid = ((pud->pud_gid != pud->pud_fsgid) ||
271 (ucred->uc_identity &&
272 (pud->pud_gid != ucred->uc_identity->mi_gid)));
274 /* check permission of setuid */
275 if (setuid && !(perm & CFS_SETUID_PERM)) {
276 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
277 pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
278 GOTO(out, rc = -EACCES);
281 /* check permission of setgid */
282 if (setgid && !(perm & CFS_SETGID_PERM)) {
283 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
284 "from %s\n", pud->pud_uid, pud->pud_gid,
285 pud->pud_fsuid, pud->pud_fsgid,
286 ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
287 GOTO(out, rc = -EACCES);
290 if (perm & CFS_SETGRP_PERM) {
291 if (pud->pud_ngroups) {
292 /* setgroups for local client */
293 ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
294 if (!ucred->uc_ginfo) {
295 CERROR("failed to alloc %d groups\n",
297 GOTO(out, rc = -ENOMEM);
300 lustre_groups_from_list(ucred->uc_ginfo,
302 lustre_groups_sort(ucred->uc_ginfo);
304 ucred->uc_ginfo = NULL;
307 ucred->uc_suppgids[0] = -1;
308 ucred->uc_suppgids[1] = -1;
309 ucred->uc_ginfo = NULL;
312 ucred->uc_uid = pud->pud_uid;
313 ucred->uc_gid = pud->pud_gid;
314 ucred->uc_fsuid = pud->pud_fsuid;
315 ucred->uc_fsgid = pud->pud_fsgid;
317 /* process root_squash here. */
318 mdt_root_squash(info, peernid);
320 /* remove fs privilege for non-root user. */
321 if (ucred->uc_fsuid && drop_fs_cap)
322 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
324 ucred->uc_cap = pud->pud_cap;
325 ucred->uc_valid = UCRED_NEW;
326 ucred_set_jobid(info, ucred);
327 ucred_set_nid(info, ucred);
328 ucred_set_audit_enabled(info, ucred);
334 if (ucred->uc_ginfo) {
335 put_group_info(ucred->uc_ginfo);
336 ucred->uc_ginfo = NULL;
338 if (ucred->uc_identity) {
339 mdt_identity_put(mdt->mdt_identity_cache,
341 ucred->uc_identity = NULL;
349 * Check whether allow the client to set supplementary group IDs or not.
351 * \param[in] info pointer to the thread context
352 * \param[in] uc pointer to the RPC user descriptor
354 * \retval true if allow to set supplementary group IDs
355 * \retval false for other cases
357 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
361 /* 1. If identity_upcall is disabled,
362 * permit local client to do anything. */
363 if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
366 /* 2. If fail to get related identities, then forbid any client to
367 * set supplementary group IDs. */
368 if (uc->uc_identity == NULL)
371 /* 3. Check the permission in the identities. */
372 perm = mdt_identity_get_perm(uc->uc_identity,
373 mdt_info_req(info)->rq_peer.nid);
374 if (perm & CFS_SETGRP_PERM)
380 int mdt_check_ucred(struct mdt_thread_info *info)
382 struct ptlrpc_request *req = mdt_info_req(info);
383 struct mdt_device *mdt = info->mti_mdt;
384 struct ptlrpc_user_desc *pud = req->rq_user_desc;
385 struct lu_ucred *ucred = mdt_ucred(info);
386 struct md_identity *identity = NULL;
387 lnet_nid_t peernid = req->rq_peer.nid;
395 LASSERT(ucred != NULL);
396 if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
399 if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
402 /* sanity check: if we use strong authentication, we expect the
403 * uid which client claimed is true */
404 if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
405 req->rq_auth_uid != pud->pud_uid) {
406 CDEBUG(D_SEC, "local client %s: auth uid %u "
407 "while client claims %u:%u/%u:%u\n",
408 libcfs_nid2str(peernid), req->rq_auth_uid,
409 pud->pud_uid, pud->pud_gid,
410 pud->pud_fsuid, pud->pud_fsgid);
414 if (is_identity_get_disabled(mdt->mdt_identity_cache))
417 identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
418 if (IS_ERR(identity)) {
419 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
422 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
428 perm = mdt_identity_get_perm(identity, peernid);
429 /* find out the setuid/setgid attempt */
430 setuid = (pud->pud_uid != pud->pud_fsuid);
431 setgid = (pud->pud_gid != pud->pud_fsgid ||
432 pud->pud_gid != identity->mi_gid);
434 /* check permission of setuid */
435 if (setuid && !(perm & CFS_SETUID_PERM)) {
436 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
437 pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
438 GOTO(out, rc = -EACCES);
441 /* check permission of setgid */
442 if (setgid && !(perm & CFS_SETGID_PERM)) {
443 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
444 "from %s\n", pud->pud_uid, pud->pud_gid,
445 pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
446 libcfs_nid2str(peernid));
447 GOTO(out, rc = -EACCES);
453 mdt_identity_put(mdt->mdt_identity_cache, identity);
457 static int old_init_ucred_common(struct mdt_thread_info *info,
458 struct lu_nodemap *nodemap,
461 struct lu_ucred *uc = mdt_ucred(info);
462 struct mdt_device *mdt = info->mti_mdt;
463 struct md_identity *identity = NULL;
465 if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
466 /* deny access before we get identity ref */
467 if (nodemap->nmf_deny_unknown)
470 uc->uc_fsuid = nodemap->nm_squash_uid;
471 uc->uc_fsgid = nodemap->nm_squash_gid;
473 uc->uc_suppgids[0] = -1;
474 uc->uc_suppgids[1] = -1;
477 if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
478 identity = mdt_identity_get(mdt->mdt_identity_cache,
480 if (IS_ERR(identity)) {
481 if (unlikely(PTR_ERR(identity) == -EREMCHG ||
482 uc->uc_cap & CFS_CAP_FS_MASK)) {
485 CDEBUG(D_SEC, "Deny access without identity: "
486 "uid %u\n", uc->uc_fsuid);
491 uc->uc_identity = identity;
493 /* process root_squash here. */
494 mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
496 /* remove fs privilege for non-root user. */
497 if (uc->uc_fsuid && drop_fs_cap)
498 uc->uc_cap &= ~CFS_CAP_FS_MASK;
499 uc->uc_valid = UCRED_OLD;
500 ucred_set_jobid(info, uc);
501 ucred_set_nid(info, uc);
502 ucred_set_audit_enabled(info, uc);
509 static int old_init_ucred(struct mdt_thread_info *info,
510 struct mdt_body *body, bool drop_fs_cap)
512 struct lu_ucred *uc = mdt_ucred(info);
513 struct lu_nodemap *nodemap;
517 nodemap = nodemap_get_from_exp(info->mti_exp);
519 RETURN(PTR_ERR(nodemap));
521 body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
522 NODEMAP_CLIENT_TO_FS, body->mbo_uid);
523 body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
524 NODEMAP_CLIENT_TO_FS, body->mbo_gid);
525 body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
526 NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
527 body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
528 NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
531 uc->uc_valid = UCRED_INVALID;
532 uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
533 uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
534 uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
535 uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
536 uc->uc_suppgids[0] = body->mbo_suppgid;
537 uc->uc_suppgids[1] = -1;
539 uc->uc_cap = body->mbo_capability;
541 rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
542 nodemap_putref(nodemap);
547 static int old_init_ucred_reint(struct mdt_thread_info *info)
549 struct lu_ucred *uc = mdt_ucred(info);
550 struct lu_nodemap *nodemap;
554 nodemap = nodemap_get_from_exp(info->mti_exp);
556 RETURN(PTR_ERR(nodemap));
560 uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
561 NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
562 uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
563 NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
565 uc->uc_valid = UCRED_INVALID;
566 uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
567 uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
570 rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */
571 nodemap_putref(nodemap);
576 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
577 struct mdt_body *body,
580 struct ptlrpc_request *req = mdt_info_req(info);
581 struct lu_ucred *uc = mdt_ucred(info);
584 if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
587 mdt_exit_ucred(info);
589 if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
590 return old_init_ucred(info, body, drop_fs_cap);
592 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
595 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
597 return __mdt_init_ucred(info, body, true);
600 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
601 * doesn't set correct fsuid explicitely, but raise capability to allow
602 * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
603 * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
604 * will fail permission check. */
605 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
606 struct mdt_body *body)
608 return __mdt_init_ucred(info, body, false);
611 int mdt_init_ucred_reint(struct mdt_thread_info *info)
613 struct ptlrpc_request *req = mdt_info_req(info);
614 struct lu_ucred *uc = mdt_ucred(info);
615 struct md_attr *ma = &info->mti_attr;
618 if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
621 /* LU-5564: for normal close request, skip permission check */
622 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
623 !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
624 uc->uc_cap |= CFS_CAP_FS_MASK;
626 mdt_exit_ucred(info);
628 if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
629 return old_init_ucred_reint(info);
631 return new_init_ucred(info, REC_INIT, NULL, true);
634 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
635 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
637 const struct lov_ost_data_v1 *lod;
638 __u32 lmm_magic = le32_to_cpu(lmm->lmm_magic);
642 if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
645 CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
646 POSTID(&lmm->lmm_oi), lmm_magic,
647 le32_to_cpu(lmm->lmm_pattern));
649 /* No support for compount layouts yet */
650 if (lmm_magic != LOV_MAGIC_V1 && lmm_magic != LOV_MAGIC_V3)
653 count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
654 CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
655 le32_to_cpu(lmm->lmm_stripe_size), count);
657 /* If it's a directory or a released file, then there are
658 * no actual objects to print, so bail out. */
659 if (valid & OBD_MD_FLDIREA ||
660 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
663 LASSERT(count <= LOV_MAX_STRIPE_COUNT);
664 for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
667 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
668 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
669 i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
673 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
675 const struct lmv_mds_md_v1 *lmm1;
678 if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
681 lmm1 = &lmv->lmv_md_v1;
683 "magic 0x%08X, master %#X stripe_count %#x hash_type %#x\n",
684 le32_to_cpu(lmm1->lmv_magic),
685 le32_to_cpu(lmm1->lmv_master_mdt_index),
686 le32_to_cpu(lmm1->lmv_stripe_count),
687 le32_to_cpu(lmm1->lmv_hash_type));
689 if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
692 for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
695 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
696 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
700 /* Shrink and/or grow reply buffers */
701 int mdt_fix_reply(struct mdt_thread_info *info)
703 struct req_capsule *pill = info->mti_pill;
704 struct mdt_body *body;
705 int md_size, md_packed = 0;
710 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
711 LASSERT(body != NULL);
713 if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
715 md_size = body->mbo_eadatasize;
719 acl_size = body->mbo_aclsize;
721 /* this replay - not send info to client */
722 if (info->mti_spec.no_create) {
727 CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
732 &RMF_ACL, or &RMF_LOGCOOKIES
733 (optional) &RMF_CAPA1,
734 (optional) &RMF_CAPA2,
735 (optional) something else
738 /* MDT_MD buffer may be bigger than packed value, let's shrink all
739 * buffers before growing it */
740 if (info->mti_big_lmm_used) {
741 /* big_lmm buffer may be used even without packing the result
742 * into reply, just for internal server needs */
743 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
744 md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
747 /* free big lmm if md_size is not needed */
748 if (md_size == 0 || md_packed == 0) {
749 info->mti_big_lmm_used = 0;
751 /* buffer must be allocated separately */
752 LASSERT(info->mti_attr.ma_lmm !=
753 req_capsule_server_get(pill, &RMF_MDT_MD));
754 req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
756 } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
757 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
760 if (info->mti_big_acl_used) {
762 info->mti_big_acl_used = 0;
764 req_capsule_shrink(pill, &RMF_ACL, 0, RCL_SERVER);
765 } else if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER)) {
766 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
767 } else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) {
768 req_capsule_shrink(pill, &RMF_LOGCOOKIES, acl_size, RCL_SERVER);
772 * Some more field should be shrinked if needed.
773 * This should be done by those who added fields to reply message.
776 /* Grow MD buffer if needed finally */
777 if (info->mti_big_lmm_used) {
780 LASSERT(md_size > md_packed);
781 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
782 md_size - md_packed);
783 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
785 /* we can't answer with proper LOV EA, drop flags,
786 * the rc is also returned so this request is
787 * considered as failed */
788 body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
789 /* don't return transno along with error */
790 lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
792 /* now we need to pack right LOV/LMV EA */
793 lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
794 if (info->mti_attr.ma_valid & MA_LOV) {
795 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
797 info->mti_attr.ma_lmm_size);
798 memcpy(lmm, info->mti_attr.ma_lmm,
799 info->mti_attr.ma_lmm_size);
800 } else if (info->mti_attr.ma_valid & MA_LMV) {
801 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
803 info->mti_attr.ma_lmv_size);
804 memcpy(lmm, info->mti_attr.ma_lmv,
805 info->mti_attr.ma_lmv_size);
809 /* update mdt_max_mdsize so clients will be aware about that */
810 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
811 info->mti_mdt->mdt_max_mdsize =
812 info->mti_attr.ma_lmm_size;
813 info->mti_big_lmm_used = 0;
816 if (info->mti_big_acl_used) {
817 CDEBUG(D_INFO, "Enlarge reply ACL buffer to %d bytes\n",
820 rc = req_capsule_server_grow(pill, &RMF_ACL, acl_size);
822 body->mbo_valid &= ~OBD_MD_FLACL;
824 void *acl = req_capsule_server_get(pill, &RMF_ACL);
826 memcpy(acl, info->mti_big_acl, acl_size);
829 info->mti_big_acl_used = 0;
836 /* if object is dying, pack the lov/llog data,
837 * parameter info->mti_attr should be valid at this point!
838 * Also implements RAoLU policy */
839 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
842 struct mdt_body *repbody = NULL;
843 const struct lu_attr *la = &ma->ma_attr;
844 struct coordinator *cdt = &info->mti_mdt->mdt_coordinator;
847 struct hsm_action_item hai = {
848 .hai_len = sizeof(hai),
849 .hai_action = HSMA_REMOVE,
850 .hai_extent.length = -1,
858 if (mdt_info_req(info) != NULL) {
859 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
860 LASSERT(repbody != NULL);
862 CDEBUG(D_INFO, "not running in a request/reply context\n");
865 if ((ma->ma_valid & MA_INODE) && repbody != NULL)
866 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
868 if (ma->ma_valid & MA_LOV) {
869 CERROR("No need in LOV EA upon unlink\n");
873 repbody->mbo_eadatasize = 0;
875 /* Only check unlinked and archived if RAoLU and upon last close */
876 if (!cdt->cdt_remove_archive_on_last_unlink ||
877 atomic_read(&mo->mot_open_count) != 0)
880 /* mdt_attr_get_complex will clear ma_valid, so check here first */
881 if ((ma->ma_valid & MA_INODE) && (ma->ma_attr.la_nlink != 0))
884 if ((ma->ma_valid & MA_HSM) && (!(ma->ma_hsm.mh_flags & HS_EXISTS)))
887 need |= (MA_INODE | MA_HSM) & ~ma->ma_valid;
889 /* ma->ma_valid is missing either MA_INODE, MA_HSM, or both,
890 * try setting them */
892 rc = mdt_attr_get_complex(info, mo, ma);
894 CERROR("%s: unable to fetch missing attributes of"
895 DFID": rc=%d\n", mdt_obd_name(info->mti_mdt),
896 PFID(mdt_object_fid(mo)), rc);
900 if (need & MA_INODE) {
901 if (ma->ma_valid & MA_INODE) {
902 if (ma->ma_attr.la_nlink != 0)
910 if (ma->ma_valid & MA_HSM) {
911 if (!(ma->ma_hsm.mh_flags & HS_EXISTS))
919 /* RAoLU policy is active, last close on file has occured,
920 * file is unlinked, file is archived, so create remove request
922 * If CDT is not running, requests will be logged for later. */
923 if (ma->ma_hsm.mh_arch_id != 0)
924 archive_id = ma->ma_hsm.mh_arch_id;
926 archive_id = cdt->cdt_default_archive_id;
928 hai.hai_fid = *mdt_object_fid(mo);
930 rc = mdt_agent_record_add(info->mti_env, info->mti_mdt, archive_id, 0,
933 CERROR("%s: unable to add HSM remove request for "DFID
934 ": rc=%d\n", mdt_obd_name(info->mti_mdt),
935 PFID(mdt_object_fid(mo)), rc);
940 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
946 if (in & MDS_ATTR_MODE)
948 if (in & MDS_ATTR_UID)
950 if (in & MDS_ATTR_GID)
952 if (in & MDS_ATTR_SIZE)
954 if (in & MDS_ATTR_BLOCKS)
956 if (in & MDS_ATTR_ATIME_SET)
958 if (in & MDS_ATTR_CTIME_SET)
960 if (in & MDS_ATTR_MTIME_SET)
962 if (in & MDS_ATTR_ATTR_FLAG)
964 if (in & MDS_ATTR_KILL_SUID)
966 if (in & MDS_ATTR_KILL_SGID)
968 if (in & MDS_ATTR_PROJID)
970 if (in & MDS_ATTR_LSIZE)
972 if (in & MDS_ATTR_LBLOCKS)
975 if (in & MDS_ATTR_FROM_OPEN)
976 rr->rr_flags |= MRF_OPEN_TRUNC;
977 if (in & MDS_ATTR_OVERRIDE)
978 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
979 if (in & MDS_ATTR_FORCE)
980 ma->ma_attr_flags |= MDS_PERM_BYPASS;
982 in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | MDS_ATTR_PROJID |
983 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
984 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
985 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
986 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
987 MDS_ATTR_FROM_OPEN | MDS_ATTR_LSIZE | MDS_ATTR_LBLOCKS |
990 CERROR("Unknown attr bits: %#llx\n", in);
996 int mdt_name_unpack(struct req_capsule *pill,
997 const struct req_msg_field *field,
999 enum mdt_name_flags flags)
1001 ln->ln_name = req_capsule_client_get(pill, field);
1002 ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
1004 if (!lu_name_is_valid(ln)) {
1011 if ((flags & MNF_FIX_ANON) &&
1012 ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
1013 /* Newer (3.x) kernels use a name of "/" for the
1014 * "anonymous" disconnected dentries from NFS
1015 * filehandle conversion. See d_obtain_alias(). */
1023 static int mdt_file_secctx_unpack(struct req_capsule *pill,
1024 const char **secctx_name,
1025 void **secctx, size_t *secctx_size)
1030 *secctx_name = NULL;
1034 if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
1035 !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
1038 name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
1043 name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1044 if (strnlen(name, name_size) != name_size - 1)
1047 if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
1048 !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
1051 *secctx_name = name;
1052 *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
1053 *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
1058 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
1060 struct lu_ucred *uc = mdt_ucred(info);
1061 struct md_attr *ma = &info->mti_attr;
1062 struct lu_attr *la = &ma->ma_attr;
1063 struct req_capsule *pill = info->mti_pill;
1064 struct mdt_reint_record *rr = &info->mti_rr;
1065 struct mdt_rec_setattr *rec;
1066 struct lu_nodemap *nodemap;
1070 CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1071 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1075 /* This prior initialization is needed for old_init_ucred_reint() */
1076 uc->uc_fsuid = rec->sa_fsuid;
1077 uc->uc_fsgid = rec->sa_fsgid;
1078 uc->uc_cap = rec->sa_cap;
1079 uc->uc_suppgids[0] = rec->sa_suppgid;
1080 uc->uc_suppgids[1] = -1;
1082 rr->rr_fid1 = &rec->sa_fid;
1083 la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
1084 la->la_mode = rec->sa_mode;
1085 la->la_flags = rec->sa_attr_flags;
1087 nodemap = nodemap_get_from_exp(info->mti_exp);
1088 if (IS_ERR(nodemap))
1089 RETURN(PTR_ERR(nodemap));
1091 la->la_uid = nodemap_map_id(nodemap, NODEMAP_UID,
1092 NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1093 la->la_gid = nodemap_map_id(nodemap, NODEMAP_GID,
1094 NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1095 la->la_projid = rec->sa_projid;
1096 nodemap_putref(nodemap);
1098 la->la_size = rec->sa_size;
1099 la->la_blocks = rec->sa_blocks;
1100 la->la_ctime = rec->sa_ctime;
1101 la->la_atime = rec->sa_atime;
1102 la->la_mtime = rec->sa_mtime;
1103 ma->ma_valid = MA_INODE;
1105 ma->ma_attr_flags |= rec->sa_bias & (MDS_CLOSE_INTENT |
1106 MDS_DATA_MODIFIED | MDS_TRUNC_KEEP_LEASE);
1110 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
1112 struct req_capsule *pill = info->mti_pill;
1113 struct mdt_ioepoch *ioepoch;
1116 if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1117 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1121 if (ioepoch == NULL)
1124 info->mti_open_handle = ioepoch->mio_open_handle;
1129 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1130 struct req_capsule *pill = info->mti_pill;
1132 if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1133 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1134 if (info->mti_dlm_req == NULL)
1141 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1143 struct mdt_reint_record *rr = &info->mti_rr;
1144 struct md_attr *ma = &info->mti_attr;
1145 struct req_capsule *pill = info->mti_pill;
1149 rc = mdt_setattr_unpack_rec(info);
1153 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1154 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1155 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1157 if (rr->rr_eadatalen > 0) {
1158 const struct lmv_user_md *lum;
1160 lum = rr->rr_eadata;
1161 /* Sigh ma_valid(from req) does not indicate whether
1162 * it will set LOV/LMV EA, so we have to check magic */
1163 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1164 ma->ma_valid |= MA_LMV;
1165 ma->ma_lmv = (void *)rr->rr_eadata;
1166 ma->ma_lmv_size = rr->rr_eadatalen;
1168 ma->ma_valid |= MA_LOV;
1169 ma->ma_lmm = (void *)rr->rr_eadata;
1170 ma->ma_lmm_size = rr->rr_eadatalen;
1175 rc = mdt_dlmreq_unpack(info);
1179 static int mdt_close_intent_unpack(struct mdt_thread_info *info)
1181 struct md_attr *ma = &info->mti_attr;
1182 struct req_capsule *pill = info->mti_pill;
1185 if (!(ma->ma_attr_flags & MDS_CLOSE_INTENT))
1188 req_capsule_extend(pill, &RQF_MDS_CLOSE_INTENT);
1190 if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1191 req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1197 int mdt_close_unpack(struct mdt_thread_info *info)
1202 rc = mdt_close_handle_unpack(info);
1206 rc = mdt_setattr_unpack_rec(info);
1210 rc = mdt_close_intent_unpack(info);
1214 RETURN(mdt_init_ucred_reint(info));
1217 static int mdt_create_unpack(struct mdt_thread_info *info)
1219 struct lu_ucred *uc = mdt_ucred(info);
1220 struct mdt_rec_create *rec;
1221 struct lu_attr *attr = &info->mti_attr.ma_attr;
1222 struct mdt_reint_record *rr = &info->mti_rr;
1223 struct req_capsule *pill = info->mti_pill;
1224 struct md_op_spec *sp = &info->mti_spec;
1229 CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1230 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1234 /* This prior initialization is needed for old_init_ucred_reint() */
1235 uc->uc_fsuid = rec->cr_fsuid;
1236 uc->uc_fsgid = rec->cr_fsgid;
1237 uc->uc_cap = rec->cr_cap;
1238 uc->uc_suppgids[0] = rec->cr_suppgid1;
1239 uc->uc_suppgids[1] = -1;
1240 uc->uc_umask = rec->cr_umask;
1242 rr->rr_fid1 = &rec->cr_fid1;
1243 rr->rr_fid2 = &rec->cr_fid2;
1244 attr->la_mode = rec->cr_mode;
1245 attr->la_rdev = rec->cr_rdev;
1246 attr->la_uid = rec->cr_fsuid;
1247 attr->la_gid = rec->cr_fsgid;
1248 attr->la_ctime = rec->cr_time;
1249 attr->la_mtime = rec->cr_time;
1250 attr->la_atime = rec->cr_time;
1251 attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1252 LA_CTIME | LA_MTIME | LA_ATIME;
1253 memset(&sp->u, 0, sizeof(sp->u));
1254 sp->sp_cr_flags = get_mrc_cr_flags(rec);
1256 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1260 if (S_ISLNK(attr->la_mode)) {
1261 const char *tgt = NULL;
1263 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1264 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1265 tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1266 sp->u.sp_symname = tgt;
1271 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1272 if (S_ISDIR(attr->la_mode) &&
1273 req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1274 sp->u.sp_ea.eadata =
1275 req_capsule_client_get(pill, &RMF_EADATA);
1276 sp->u.sp_ea.eadatalen =
1277 req_capsule_get_size(pill, &RMF_EADATA,
1279 sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1283 rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1284 &sp->sp_cr_file_secctx,
1285 &sp->sp_cr_file_secctx_size);
1289 rc = mdt_dlmreq_unpack(info);
1293 static int mdt_link_unpack(struct mdt_thread_info *info)
1295 struct lu_ucred *uc = mdt_ucred(info);
1296 struct mdt_rec_link *rec;
1297 struct lu_attr *attr = &info->mti_attr.ma_attr;
1298 struct mdt_reint_record *rr = &info->mti_rr;
1299 struct req_capsule *pill = info->mti_pill;
1304 CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1305 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1309 /* This prior initialization is needed for old_init_ucred_reint() */
1310 uc->uc_fsuid = rec->lk_fsuid;
1311 uc->uc_fsgid = rec->lk_fsgid;
1312 uc->uc_cap = rec->lk_cap;
1313 uc->uc_suppgids[0] = rec->lk_suppgid1;
1314 uc->uc_suppgids[1] = rec->lk_suppgid2;
1316 attr->la_uid = rec->lk_fsuid;
1317 attr->la_gid = rec->lk_fsgid;
1318 rr->rr_fid1 = &rec->lk_fid1;
1319 rr->rr_fid2 = &rec->lk_fid2;
1320 attr->la_ctime = rec->lk_time;
1321 attr->la_mtime = rec->lk_time;
1322 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1324 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1328 rc = mdt_dlmreq_unpack(info);
1333 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1335 struct lu_ucred *uc = mdt_ucred(info);
1336 struct mdt_rec_unlink *rec;
1337 struct lu_attr *attr = &info->mti_attr.ma_attr;
1338 struct mdt_reint_record *rr = &info->mti_rr;
1339 struct req_capsule *pill = info->mti_pill;
1344 CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1345 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1349 /* This prior initialization is needed for old_init_ucred_reint() */
1350 uc->uc_fsuid = rec->ul_fsuid;
1351 uc->uc_fsgid = rec->ul_fsgid;
1352 uc->uc_cap = rec->ul_cap;
1353 uc->uc_suppgids[0] = rec->ul_suppgid1;
1354 uc->uc_suppgids[1] = -1;
1356 attr->la_uid = rec->ul_fsuid;
1357 attr->la_gid = rec->ul_fsgid;
1358 rr->rr_fid1 = &rec->ul_fid1;
1359 rr->rr_fid2 = &rec->ul_fid2;
1360 attr->la_ctime = rec->ul_time;
1361 attr->la_mtime = rec->ul_time;
1362 attr->la_mode = rec->ul_mode;
1363 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1365 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1369 info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1371 rc = mdt_dlmreq_unpack(info);
1375 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1377 info->mti_spec.sp_rm_entry = 1;
1378 return mdt_unlink_unpack(info);
1381 static int mdt_rename_unpack(struct mdt_thread_info *info)
1383 struct lu_ucred *uc = mdt_ucred(info);
1384 struct mdt_rec_rename *rec;
1385 struct lu_attr *attr = &info->mti_attr.ma_attr;
1386 struct mdt_reint_record *rr = &info->mti_rr;
1387 struct req_capsule *pill = info->mti_pill;
1388 struct md_op_spec *spec = &info->mti_spec;
1393 CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1394 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1398 /* This prior initialization is needed for old_init_ucred_reint() */
1399 uc->uc_fsuid = rec->rn_fsuid;
1400 uc->uc_fsgid = rec->rn_fsgid;
1401 uc->uc_cap = rec->rn_cap;
1402 uc->uc_suppgids[0] = rec->rn_suppgid1;
1403 uc->uc_suppgids[1] = rec->rn_suppgid2;
1405 attr->la_uid = rec->rn_fsuid;
1406 attr->la_gid = rec->rn_fsgid;
1407 rr->rr_fid1 = &rec->rn_fid1;
1408 rr->rr_fid2 = &rec->rn_fid2;
1409 attr->la_ctime = rec->rn_time;
1410 attr->la_mtime = rec->rn_time;
1411 /* rename_tgt contains the mode already */
1412 attr->la_mode = rec->rn_mode;
1413 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1415 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1419 rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1423 spec->no_create = !!req_is_replay(mdt_info_req(info));
1425 rc = mdt_dlmreq_unpack(info);
1430 static int mdt_migrate_unpack(struct mdt_thread_info *info)
1432 struct lu_ucred *uc = mdt_ucred(info);
1433 struct mdt_rec_rename *rec;
1434 struct lu_attr *attr = &info->mti_attr.ma_attr;
1435 struct mdt_reint_record *rr = &info->mti_rr;
1436 struct req_capsule *pill = info->mti_pill;
1437 struct md_op_spec *spec = &info->mti_spec;
1442 CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1443 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1447 /* This prior initialization is needed for old_init_ucred_reint() */
1448 uc->uc_fsuid = rec->rn_fsuid;
1449 uc->uc_fsgid = rec->rn_fsgid;
1450 uc->uc_cap = rec->rn_cap;
1451 uc->uc_suppgids[0] = rec->rn_suppgid1;
1452 uc->uc_suppgids[1] = rec->rn_suppgid2;
1454 attr->la_uid = rec->rn_fsuid;
1455 attr->la_gid = rec->rn_fsgid;
1456 rr->rr_fid1 = &rec->rn_fid1;
1457 rr->rr_fid2 = &rec->rn_fid2;
1458 attr->la_ctime = rec->rn_time;
1459 attr->la_mtime = rec->rn_time;
1460 /* rename_tgt contains the mode already */
1461 attr->la_mode = rec->rn_mode;
1462 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1464 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1468 if (rec->rn_bias & MDS_CLOSE_MIGRATE) {
1469 rc = mdt_close_handle_unpack(info);
1473 spec->sp_migrate_close = 1;
1476 /* lustre version > 2.11 migration packs lum */
1477 if (req_capsule_has_field(pill, &RMF_EADATA, RCL_CLIENT)) {
1478 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1479 rr->rr_eadatalen = req_capsule_get_size(pill,
1482 if (rr->rr_eadatalen > 0) {
1483 rr->rr_eadata = req_capsule_client_get(pill,
1485 spec->u.sp_ea.eadatalen = rr->rr_eadatalen;
1486 spec->u.sp_ea.eadata = rr->rr_eadata;
1487 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1490 /* old client doesn't provide lum. */
1491 RETURN(-EOPNOTSUPP);
1495 spec->no_create = !!req_is_replay(mdt_info_req(info));
1497 rc = mdt_dlmreq_unpack(info);
1503 * please see comment above LOV_MAGIC_V1_DEFINED
1505 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata)
1507 struct lov_user_md_v1 *v1 = eadata;
1511 if (unlikely(req_is_replay(mdt_info_req(info)))) {
1512 if ((v1->lmm_magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
1513 v1->lmm_magic |= LOV_MAGIC_DEFINED;
1514 else if ((v1->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
1515 __swab32(LOV_MAGIC_MAGIC))
1516 v1->lmm_magic |= __swab32(LOV_MAGIC_DEFINED);
1520 static int mdt_open_unpack(struct mdt_thread_info *info)
1522 struct lu_ucred *uc = mdt_ucred(info);
1523 struct mdt_rec_create *rec;
1524 struct lu_attr *attr = &info->mti_attr.ma_attr;
1525 struct req_capsule *pill = info->mti_pill;
1526 struct mdt_reint_record *rr = &info->mti_rr;
1527 struct ptlrpc_request *req = mdt_info_req(info);
1528 struct md_op_spec *sp = &info->mti_spec;
1532 CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1533 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1537 /* This prior initialization is needed for old_init_ucred_reint() */
1538 uc->uc_fsuid = rec->cr_fsuid;
1539 uc->uc_fsgid = rec->cr_fsgid;
1540 uc->uc_cap = rec->cr_cap;
1541 uc->uc_suppgids[0] = rec->cr_suppgid1;
1542 uc->uc_suppgids[1] = rec->cr_suppgid2;
1543 uc->uc_umask = rec->cr_umask;
1545 rr->rr_fid1 = &rec->cr_fid1;
1546 rr->rr_fid2 = &rec->cr_fid2;
1547 rr->rr_open_handle = &rec->cr_open_handle_old;
1548 attr->la_mode = rec->cr_mode;
1549 attr->la_rdev = rec->cr_rdev;
1550 attr->la_uid = rec->cr_fsuid;
1551 attr->la_gid = rec->cr_fsgid;
1552 attr->la_ctime = rec->cr_time;
1553 attr->la_mtime = rec->cr_time;
1554 attr->la_atime = rec->cr_time;
1555 attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID |
1556 LA_CTIME | LA_MTIME | LA_ATIME;
1557 memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1558 info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1559 /* Do not trigger ASSERTION if client miss to set such flags. */
1560 if (unlikely(info->mti_spec.sp_cr_flags == 0))
1563 info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1565 mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1567 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1568 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1570 if (rr->rr_eadatalen > 0) {
1571 rr->rr_eadata = req_capsule_client_get(pill,
1573 sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1574 sp->u.sp_ea.eadata = rr->rr_eadata;
1575 sp->no_create = !!req_is_replay(req);
1576 mdt_fix_lov_magic(info, rr->rr_eadata);
1580 * Client default md_size may be 0 right after client start,
1581 * until all osc are connected, set here just some reasonable
1582 * value to prevent misbehavior.
1584 if (rr->rr_eadatalen == 0 &&
1585 !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1586 rr->rr_eadatalen = MIN_MD_SIZE;
1589 rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1590 &sp->sp_cr_file_secctx,
1591 &sp->sp_cr_file_secctx_size);
1596 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1598 struct mdt_reint_record *rr = &info->mti_rr;
1599 struct lu_ucred *uc = mdt_ucred(info);
1600 struct lu_attr *attr = &info->mti_attr.ma_attr;
1601 struct req_capsule *pill = info->mti_pill;
1602 struct mdt_rec_setxattr *rec;
1607 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1608 sizeof(struct mdt_rec_reint));
1610 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1614 /* This prior initialization is needed for old_init_ucred_reint() */
1615 uc->uc_fsuid = rec->sx_fsuid;
1616 uc->uc_fsgid = rec->sx_fsgid;
1617 uc->uc_cap = rec->sx_cap;
1618 uc->uc_suppgids[0] = rec->sx_suppgid1;
1619 uc->uc_suppgids[1] = -1;
1621 rr->rr_opcode = rec->sx_opcode;
1622 rr->rr_fid1 = &rec->sx_fid;
1623 attr->la_valid = rec->sx_valid;
1624 attr->la_ctime = rec->sx_time;
1625 attr->la_size = rec->sx_size;
1626 attr->la_flags = rec->sx_flags;
1628 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1632 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1633 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1635 if (rr->rr_eadatalen > 0) {
1636 rr->rr_eadata = req_capsule_client_get(pill,
1638 if (rr->rr_eadata == NULL)
1641 rr->rr_eadata = NULL;
1643 } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1644 CDEBUG(D_INFO, "no xattr data supplied\n");
1648 if (mdt_dlmreq_unpack(info) < 0)
1654 static int mdt_resync_unpack(struct mdt_thread_info *info)
1656 struct req_capsule *pill = info->mti_pill;
1657 struct mdt_reint_record *rr = &info->mti_rr;
1658 struct lu_ucred *uc = mdt_ucred(info);
1659 struct mdt_rec_resync *rec;
1662 CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1663 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1667 /* This prior initialization is needed for old_init_ucred_reint() */
1668 uc->uc_fsuid = rec->rs_fsuid;
1669 uc->uc_fsgid = rec->rs_fsgid;
1670 uc->uc_cap = rec->rs_cap;
1672 rr->rr_fid1 = &rec->rs_fid;
1673 rr->rr_mirror_id = rec->rs_mirror_id;
1675 /* cookie doesn't need to be swapped but it has been swapped
1676 * in lustre_swab_mdt_rec_reint() as rr_mtime, so here it needs
1678 if (ptlrpc_req_need_swab(mdt_info_req(info)))
1679 __swab64s(&rec->rs_lease_handle.cookie);
1680 rr->rr_lease_handle = &rec->rs_lease_handle;
1682 RETURN(mdt_dlmreq_unpack(info));
1685 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1687 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1688 [REINT_SETATTR] = mdt_setattr_unpack,
1689 [REINT_CREATE] = mdt_create_unpack,
1690 [REINT_LINK] = mdt_link_unpack,
1691 [REINT_UNLINK] = mdt_unlink_unpack,
1692 [REINT_RENAME] = mdt_rename_unpack,
1693 [REINT_OPEN] = mdt_open_unpack,
1694 [REINT_SETXATTR] = mdt_setxattr_unpack,
1695 [REINT_RMENTRY] = mdt_rmentry_unpack,
1696 [REINT_MIGRATE] = mdt_migrate_unpack,
1697 [REINT_RESYNC] = mdt_resync_unpack,
1700 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1705 memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1706 if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1707 info->mti_rr.rr_opcode = op;
1708 rc = mdt_reint_unpackers[op](info);
1710 CERROR("Unexpected opcode %d\n", op);