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 #include <linux/uidgid.h>
50 #include "mdt_internal.h"
51 #include <uapi/linux/lnet/nidstr.h>
52 #include <lustre_nodemap.h>
54 typedef enum ucred_init_type {
60 static __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
62 return (__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32);
65 void mdt_exit_ucred(struct mdt_thread_info *info)
67 struct lu_ucred *uc = mdt_ucred(info);
68 struct mdt_device *mdt = info->mti_mdt;
71 if (uc->uc_valid != UCRED_INIT) {
72 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
74 put_group_info(uc->uc_ginfo);
77 if (uc->uc_identity) {
78 mdt_identity_put(mdt->mdt_identity_cache,
80 uc->uc_identity = NULL;
82 uc->uc_valid = UCRED_INIT;
86 static int match_nosquash_list(struct spinlock *rsi_lock,
87 struct list_head *nidlist,
93 rc = cfs_match_nid(peernid, nidlist);
94 spin_unlock(rsi_lock);
98 /* root_squash for inter-MDS operations */
99 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
101 struct lu_ucred *ucred = mdt_ucred(info);
102 struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
105 LASSERT(ucred != NULL);
106 if (!squash->rsi_uid || ucred->uc_fsuid)
109 if (match_nosquash_list(&squash->rsi_lock,
110 &squash->rsi_nosquash_nids,
112 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
113 libcfs_nid2str(peernid));
117 CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
118 libcfs_nid2str(peernid),
119 ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
120 squash->rsi_uid, squash->rsi_gid, 0);
122 ucred->uc_fsuid = squash->rsi_uid;
123 ucred->uc_fsgid = squash->rsi_gid;
125 ucred->uc_suppgids[0] = -1;
126 ucred->uc_suppgids[1] = -1;
131 static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
133 struct ptlrpc_request *req = mdt_info_req(info);
134 const char *jobid = mdt_req_get_jobid(req);
136 /* set jobid if specified. */
138 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
140 uc->uc_jobid[0] = '\0';
143 static void ucred_set_nid(struct mdt_thread_info *info, struct lu_ucred *uc)
145 if (info && info->mti_exp && info->mti_exp->exp_connection)
146 uc->uc_nid = info->mti_exp->exp_connection->c_peer.nid;
148 uc->uc_nid = LNET_NID_ANY;
151 static void ucred_set_audit_enabled(struct mdt_thread_info *info,
154 struct lu_nodemap *nodemap = NULL;
157 if (info && info->mti_exp) {
158 nodemap = nodemap_get_from_exp(info->mti_exp);
159 if (nodemap && !IS_ERR(nodemap)) {
160 audit = nodemap->nmf_enable_audit;
161 nodemap_putref(nodemap);
165 uc->uc_enable_audit = audit;
168 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
171 struct ptlrpc_request *req = mdt_info_req(info);
172 struct mdt_device *mdt = info->mti_mdt;
173 struct ptlrpc_user_desc *pud = req->rq_user_desc;
174 struct lu_ucred *ucred = mdt_ucred(info);
175 struct lu_nodemap *nodemap;
176 lnet_nid_t peernid = req->rq_peer.nid;
180 bool is_nm_gid_squashed = false;
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_suppgids[0] = -1;
218 ucred->uc_suppgids[1] = -1;
221 if (nodemap && ucred->uc_o_gid == nodemap->nm_squash_gid)
222 is_nm_gid_squashed = true;
224 nodemap_putref(nodemap);
226 if (type == BODY_INIT) {
227 struct mdt_body *body = (struct mdt_body *)buf;
229 ucred->uc_suppgids[0] = body->mbo_suppgid;
230 ucred->uc_suppgids[1] = -1;
233 if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
234 req->rq_auth_uid != pud->pud_uid) {
235 CDEBUG(D_SEC, "local client %s: auth uid %u "
236 "while client claims %u:%u/%u:%u\n",
237 libcfs_nid2str(peernid), req->rq_auth_uid,
238 pud->pud_uid, pud->pud_gid,
239 pud->pud_fsuid, pud->pud_fsgid);
243 if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
244 ucred->uc_identity = NULL;
245 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
247 struct md_identity *identity;
249 identity = mdt_identity_get(mdt->mdt_identity_cache,
251 if (IS_ERR(identity)) {
252 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
253 ucred->uc_identity = NULL;
254 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
258 "Deny access without identity: uid %u\n",
263 ucred->uc_identity = identity;
264 perm = mdt_identity_get_perm(ucred->uc_identity,
269 /* find out the setuid/setgid attempt */
270 setuid = (pud->pud_uid != pud->pud_fsuid);
271 setgid = ((pud->pud_gid != pud->pud_fsgid) ||
272 (ucred->uc_identity &&
273 (pud->pud_gid != ucred->uc_identity->mi_gid)));
275 /* check permission of setuid */
276 if (setuid && !(perm & CFS_SETUID_PERM)) {
277 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
278 pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
279 GOTO(out, rc = -EACCES);
282 /* check permission of setgid */
283 if (setgid && !(perm & CFS_SETGID_PERM)) {
284 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
285 "from %s\n", pud->pud_uid, pud->pud_gid,
286 pud->pud_fsuid, pud->pud_fsgid,
287 ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
288 GOTO(out, rc = -EACCES);
291 if (perm & CFS_SETGRP_PERM) {
292 /* only set groups if GID is not squashed */
293 if (pud->pud_ngroups && !is_nm_gid_squashed) {
294 /* setgroups for local client */
295 ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
296 if (!ucred->uc_ginfo) {
297 CERROR("failed to alloc %d groups\n",
299 GOTO(out, rc = -ENOMEM);
302 lustre_groups_from_list(ucred->uc_ginfo,
304 lustre_groups_sort(ucred->uc_ginfo);
306 ucred->uc_suppgids[0] = -1;
307 ucred->uc_suppgids[1] = -1;
308 ucred->uc_ginfo = NULL;
311 ucred->uc_suppgids[0] = -1;
312 ucred->uc_suppgids[1] = -1;
313 ucred->uc_ginfo = NULL;
316 ucred->uc_uid = pud->pud_uid;
317 ucred->uc_gid = pud->pud_gid;
319 if (nodemap && ucred->uc_o_uid == nodemap->nm_squash_uid) {
322 ucred->uc_cap = pud->pud_cap;
324 ucred->uc_fsuid = pud->pud_fsuid;
325 ucred->uc_fsgid = pud->pud_fsgid;
327 /* process root_squash here. */
328 mdt_root_squash(info, peernid);
330 ucred->uc_valid = UCRED_NEW;
331 ucred_set_jobid(info, ucred);
332 ucred_set_nid(info, ucred);
333 ucred_set_audit_enabled(info, ucred);
339 if (ucred->uc_ginfo) {
340 put_group_info(ucred->uc_ginfo);
341 ucred->uc_ginfo = NULL;
343 if (ucred->uc_identity) {
344 mdt_identity_put(mdt->mdt_identity_cache,
346 ucred->uc_identity = NULL;
354 * Check whether allow the client to set supplementary group IDs or not.
356 * \param[in] info pointer to the thread context
357 * \param[in] uc pointer to the RPC user descriptor
359 * \retval true if allow to set supplementary group IDs
360 * \retval false for other cases
362 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
366 /* 1. If identity_upcall is disabled,
367 * permit local client to do anything. */
368 if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
371 /* 2. If fail to get related identities, then forbid any client to
372 * set supplementary group IDs. */
373 if (uc->uc_identity == NULL)
376 /* 3. Check the permission in the identities. */
377 perm = mdt_identity_get_perm(uc->uc_identity,
378 mdt_info_req(info)->rq_peer.nid);
379 if (perm & CFS_SETGRP_PERM)
385 int mdt_check_ucred(struct mdt_thread_info *info)
387 struct ptlrpc_request *req = mdt_info_req(info);
388 struct mdt_device *mdt = info->mti_mdt;
389 struct ptlrpc_user_desc *pud = req->rq_user_desc;
390 struct lu_ucred *ucred = mdt_ucred(info);
391 struct md_identity *identity = NULL;
392 lnet_nid_t peernid = req->rq_peer.nid;
400 LASSERT(ucred != NULL);
401 if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
404 if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
407 /* sanity check: if we use strong authentication, we expect the
408 * uid which client claimed is true */
409 if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
410 req->rq_auth_uid != pud->pud_uid) {
411 CDEBUG(D_SEC, "local client %s: auth uid %u "
412 "while client claims %u:%u/%u:%u\n",
413 libcfs_nid2str(peernid), req->rq_auth_uid,
414 pud->pud_uid, pud->pud_gid,
415 pud->pud_fsuid, pud->pud_fsgid);
419 if (is_identity_get_disabled(mdt->mdt_identity_cache))
422 identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
423 if (IS_ERR(identity)) {
424 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
427 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
433 perm = mdt_identity_get_perm(identity, peernid);
434 /* find out the setuid/setgid attempt */
435 setuid = (pud->pud_uid != pud->pud_fsuid);
436 setgid = (pud->pud_gid != pud->pud_fsgid ||
437 pud->pud_gid != identity->mi_gid);
439 /* check permission of setuid */
440 if (setuid && !(perm & CFS_SETUID_PERM)) {
441 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
442 pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
443 GOTO(out, rc = -EACCES);
446 /* check permission of setgid */
447 if (setgid && !(perm & CFS_SETGID_PERM)) {
448 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
449 "from %s\n", pud->pud_uid, pud->pud_gid,
450 pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
451 libcfs_nid2str(peernid));
452 GOTO(out, rc = -EACCES);
458 mdt_identity_put(mdt->mdt_identity_cache, identity);
462 static int old_init_ucred_common(struct mdt_thread_info *info,
463 struct lu_nodemap *nodemap)
465 struct lu_ucred *uc = mdt_ucred(info);
466 struct mdt_device *mdt = info->mti_mdt;
467 struct md_identity *identity = NULL;
469 if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
470 /* deny access before we get identity ref */
471 if (nodemap->nmf_deny_unknown)
475 uc->uc_suppgids[0] = -1;
476 uc->uc_suppgids[1] = -1;
479 if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
480 identity = mdt_identity_get(mdt->mdt_identity_cache,
482 if (IS_ERR(identity)) {
483 if (unlikely(PTR_ERR(identity) == -EREMCHG ||
484 uc->uc_cap & CFS_CAP_FS_MASK)) {
487 CDEBUG(D_SEC, "Deny access without identity: "
488 "uid %u\n", uc->uc_fsuid);
493 uc->uc_identity = identity;
495 /* process root_squash here. */
496 mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
498 uc->uc_valid = UCRED_OLD;
499 ucred_set_jobid(info, uc);
500 ucred_set_nid(info, uc);
501 ucred_set_audit_enabled(info, uc);
508 static int old_init_ucred(struct mdt_thread_info *info,
509 struct mdt_body *body)
511 struct lu_ucred *uc = mdt_ucred(info);
512 struct lu_nodemap *nodemap;
516 nodemap = nodemap_get_from_exp(info->mti_exp);
518 RETURN(PTR_ERR(nodemap));
520 body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
521 NODEMAP_CLIENT_TO_FS, body->mbo_uid);
522 body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
523 NODEMAP_CLIENT_TO_FS, body->mbo_gid);
524 body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
525 NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
526 body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
527 NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
530 uc->uc_valid = UCRED_INVALID;
531 uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
532 uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
533 uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
534 uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
535 uc->uc_suppgids[0] = body->mbo_suppgid;
536 uc->uc_suppgids[1] = -1;
538 uc->uc_cap = body->mbo_capability;
540 rc = old_init_ucred_common(info, nodemap);
541 nodemap_putref(nodemap);
546 static int old_init_ucred_reint(struct mdt_thread_info *info)
548 struct lu_ucred *uc = mdt_ucred(info);
549 struct lu_nodemap *nodemap;
553 nodemap = nodemap_get_from_exp(info->mti_exp);
555 RETURN(PTR_ERR(nodemap));
559 uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
560 NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
561 uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
562 NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
564 uc->uc_valid = UCRED_INVALID;
565 uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
566 uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
569 rc = old_init_ucred_common(info, nodemap);
570 nodemap_putref(nodemap);
575 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
576 struct mdt_body *body)
578 struct ptlrpc_request *req = mdt_info_req(info);
579 struct lu_ucred *uc = mdt_ucred(info);
582 if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
585 mdt_exit_ucred(info);
587 if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
588 return old_init_ucred(info, body);
590 return new_init_ucred(info, BODY_INIT, body);
593 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
595 return __mdt_init_ucred(info, body);
598 int mdt_init_ucred_reint(struct mdt_thread_info *info)
600 struct ptlrpc_request *req = mdt_info_req(info);
601 struct lu_ucred *uc = mdt_ucred(info);
602 struct md_attr *ma = &info->mti_attr;
605 if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
608 /* LU-5564: for normal close request, skip permission check */
609 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
610 !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
611 uc->uc_cap |= CFS_CAP_FS_MASK;
613 mdt_exit_ucred(info);
615 if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
616 return old_init_ucred_reint(info);
618 return new_init_ucred(info, REC_INIT, NULL);
621 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
622 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
624 const struct lov_ost_data_v1 *lod;
625 __u32 lmm_magic = le32_to_cpu(lmm->lmm_magic);
629 if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
632 CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
633 POSTID(&lmm->lmm_oi), lmm_magic,
634 le32_to_cpu(lmm->lmm_pattern));
636 /* No support for compount layouts yet */
637 if (lmm_magic != LOV_MAGIC_V1 && lmm_magic != LOV_MAGIC_V3)
640 count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
641 CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
642 le32_to_cpu(lmm->lmm_stripe_size), count);
644 /* If it's a directory or a released file, then there are
645 * no actual objects to print, so bail out. */
646 if (valid & OBD_MD_FLDIREA ||
647 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
650 LASSERT(count <= LOV_MAX_STRIPE_COUNT);
651 for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
654 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
655 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
656 i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
660 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
662 const struct lmv_mds_md_v1 *lmm1;
663 const struct lmv_foreign_md *lfm;
666 if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
669 /* foreign LMV case */
670 lfm = &lmv->lmv_foreign_md;
671 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
673 "foreign magic 0x%08X, length %u, type %u, flags %u, value '%.*s'\n",
674 le32_to_cpu(lfm->lfm_magic),
675 le32_to_cpu(lfm->lfm_length),
676 le32_to_cpu(lfm->lfm_type),
677 le32_to_cpu(lfm->lfm_flags),
678 le32_to_cpu(lfm->lfm_length), lfm->lfm_value);
682 lmm1 = &lmv->lmv_md_v1;
684 "magic 0x%08X, master %#X stripe_count %#x hash_type %#x\n",
685 le32_to_cpu(lmm1->lmv_magic),
686 le32_to_cpu(lmm1->lmv_master_mdt_index),
687 le32_to_cpu(lmm1->lmv_stripe_count),
688 le32_to_cpu(lmm1->lmv_hash_type));
690 if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
693 for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
696 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
697 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
701 /* Shrink and/or grow reply buffers */
702 int mdt_fix_reply(struct mdt_thread_info *info)
704 struct req_capsule *pill = info->mti_pill;
705 struct mdt_body *body;
706 int md_size, md_packed = 0;
711 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
712 LASSERT(body != NULL);
714 if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
716 md_size = body->mbo_eadatasize;
720 acl_size = body->mbo_aclsize;
722 /* this replay - not send info to client */
723 if (info->mti_spec.no_create) {
728 CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
733 &RMF_ACL, or &RMF_LOGCOOKIES
734 (optional) &RMF_CAPA1,
735 (optional) &RMF_CAPA2,
736 (optional) something else
739 /* MDT_MD buffer may be bigger than packed value, let's shrink all
740 * buffers before growing it */
741 if (info->mti_big_lmm_used) {
742 /* big_lmm buffer may be used even without packing the result
743 * into reply, just for internal server needs */
744 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
745 md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
748 /* free big lmm if md_size is not needed */
749 if (md_size == 0 || md_packed == 0) {
750 info->mti_big_lmm_used = 0;
752 /* buffer must be allocated separately */
753 LASSERT(info->mti_attr.ma_lmm !=
754 req_capsule_server_get(pill, &RMF_MDT_MD));
755 req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
757 } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
758 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
761 if (info->mti_big_acl_used) {
763 info->mti_big_acl_used = 0;
765 req_capsule_shrink(pill, &RMF_ACL, 0, RCL_SERVER);
766 } else if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER)) {
767 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
768 } else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) {
769 req_capsule_shrink(pill, &RMF_LOGCOOKIES, acl_size, RCL_SERVER);
772 /* Shrink optional SECCTX buffer if it is not used */
773 if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
774 req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0 &&
775 !(body->mbo_valid & OBD_MD_SECCTX))
776 req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0, RCL_SERVER);
778 /* Shrink optional ENCCTX buffer if it is not used */
779 if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
780 req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0 &&
781 !(body->mbo_valid & OBD_MD_ENCCTX))
782 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0, RCL_SERVER);
785 * Some more field should be shrinked if needed.
786 * This should be done by those who added fields to reply message.
789 /* Grow MD buffer if needed finally */
790 if (info->mti_big_lmm_used) {
793 LASSERT(md_size > md_packed);
794 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
795 md_size - md_packed);
796 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
798 /* we can't answer with proper LOV EA, drop flags,
799 * the rc is also returned so this request is
800 * considered as failed */
801 body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
802 /* don't return transno along with error */
803 lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
805 /* now we need to pack right LOV/LMV EA */
806 lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
807 if (info->mti_attr.ma_valid & MA_LOV) {
808 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
810 info->mti_attr.ma_lmm_size);
811 memcpy(lmm, info->mti_attr.ma_lmm,
812 info->mti_attr.ma_lmm_size);
813 } else if (info->mti_attr.ma_valid & MA_LMV) {
814 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
816 info->mti_attr.ma_lmv_size);
817 memcpy(lmm, info->mti_attr.ma_lmv,
818 info->mti_attr.ma_lmv_size);
822 /* update mdt_max_mdsize so clients will be aware about that */
823 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
824 info->mti_mdt->mdt_max_mdsize =
825 info->mti_attr.ma_lmm_size;
826 info->mti_big_lmm_used = 0;
829 if (info->mti_big_acl_used) {
830 CDEBUG(D_INFO, "Enlarge reply ACL buffer to %d bytes\n",
833 rc = req_capsule_server_grow(pill, &RMF_ACL, acl_size);
835 body->mbo_valid &= ~OBD_MD_FLACL;
837 void *acl = req_capsule_server_get(pill, &RMF_ACL);
839 memcpy(acl, info->mti_big_acl, acl_size);
842 info->mti_big_acl_used = 0;
849 /* if object is dying, pack the lov/llog data,
850 * parameter info->mti_attr should be valid at this point!
851 * Also implements RAoLU policy */
852 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
855 struct mdt_body *repbody = NULL;
856 const struct lu_attr *la = &ma->ma_attr;
857 struct coordinator *cdt = &info->mti_mdt->mdt_coordinator;
860 struct hsm_action_item hai = {
861 .hai_len = sizeof(hai),
862 .hai_action = HSMA_REMOVE,
863 .hai_extent.length = -1,
871 if (mdt_info_req(info) != NULL) {
872 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
873 LASSERT(repbody != NULL);
875 CDEBUG(D_INFO, "not running in a request/reply context\n");
878 if ((ma->ma_valid & MA_INODE) && repbody != NULL)
879 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
881 if (ma->ma_valid & MA_LOV) {
882 CERROR("No need in LOV EA upon unlink\n");
886 repbody->mbo_eadatasize = 0;
888 /* Only check unlinked and archived if RAoLU and upon last close */
889 if (!cdt->cdt_remove_archive_on_last_unlink ||
890 atomic_read(&mo->mot_open_count) != 0)
893 /* mdt_attr_get_complex will clear ma_valid, so check here first */
894 if ((ma->ma_valid & MA_INODE) && (ma->ma_attr.la_nlink != 0))
897 if ((ma->ma_valid & MA_HSM) && (!(ma->ma_hsm.mh_flags & HS_EXISTS)))
900 need |= (MA_INODE | MA_HSM) & ~ma->ma_valid;
902 /* ma->ma_valid is missing either MA_INODE, MA_HSM, or both,
903 * try setting them */
905 rc = mdt_attr_get_complex(info, mo, ma);
907 CERROR("%s: unable to fetch missing attributes of"
908 DFID": rc=%d\n", mdt_obd_name(info->mti_mdt),
909 PFID(mdt_object_fid(mo)), rc);
913 if (need & MA_INODE) {
914 if (ma->ma_valid & MA_INODE) {
915 if (ma->ma_attr.la_nlink != 0)
923 if (ma->ma_valid & MA_HSM) {
924 if (!(ma->ma_hsm.mh_flags & HS_EXISTS))
932 /* RAoLU policy is active, last close on file has occured,
933 * file is unlinked, file is archived, so create remove request
935 * If CDT is not running, requests will be logged for later. */
936 if (ma->ma_hsm.mh_arch_id != 0)
937 archive_id = ma->ma_hsm.mh_arch_id;
939 archive_id = cdt->cdt_default_archive_id;
941 hai.hai_fid = *mdt_object_fid(mo);
943 rc = mdt_agent_record_add(info->mti_env, info->mti_mdt, archive_id, 0,
946 CERROR("%s: unable to add HSM remove request for "DFID
947 ": rc=%d\n", mdt_obd_name(info->mti_mdt),
948 PFID(mdt_object_fid(mo)), rc);
953 static __u64 mdt_attr_valid_xlate(enum mds_attr_flags in,
954 struct mdt_reint_record *rr,
960 if (in & MDS_ATTR_MODE)
962 if (in & MDS_ATTR_UID)
964 if (in & MDS_ATTR_GID)
966 if (in & MDS_ATTR_SIZE)
968 if (in & MDS_ATTR_BLOCKS)
970 if (in & MDS_ATTR_ATIME_SET)
972 if (in & MDS_ATTR_CTIME_SET)
974 if (in & MDS_ATTR_MTIME_SET)
976 if (in & MDS_ATTR_ATTR_FLAG)
978 if (in & MDS_ATTR_KILL_SUID)
980 if (in & MDS_ATTR_KILL_SGID)
982 if (in & MDS_ATTR_PROJID)
984 if (in & MDS_ATTR_LSIZE)
986 if (in & MDS_ATTR_LBLOCKS)
989 if (in & MDS_ATTR_FROM_OPEN)
990 rr->rr_flags |= MRF_OPEN_TRUNC;
991 if (in & MDS_ATTR_OVERRIDE)
992 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
993 if (in & MDS_ATTR_FORCE)
994 ma->ma_attr_flags |= MDS_PERM_BYPASS;
996 in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | MDS_ATTR_PROJID |
997 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
998 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
999 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
1000 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
1001 MDS_ATTR_FROM_OPEN | MDS_ATTR_LSIZE | MDS_ATTR_LBLOCKS |
1004 CDEBUG(D_INFO, "Unknown attr bits: %#llx\n", (u64)in);
1011 int mdt_name_unpack(struct req_capsule *pill,
1012 const struct req_msg_field *field,
1014 enum mdt_name_flags flags)
1016 ln->ln_name = req_capsule_client_get(pill, field);
1017 ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
1019 if (!lu_name_is_valid(ln)) {
1026 if ((flags & MNF_FIX_ANON) &&
1027 ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
1028 /* Newer (3.x) kernels use a name of "/" for the
1029 * "anonymous" disconnected dentries from NFS
1030 * filehandle conversion. See d_obtain_alias(). */
1038 static int mdt_file_secctx_unpack(struct req_capsule *pill,
1039 const char **secctx_name,
1040 void **secctx, size_t *secctx_size)
1045 *secctx_name = NULL;
1049 if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
1050 !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
1053 name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
1058 if (name_size > XATTR_NAME_MAX + 1)
1061 name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1062 if (strnlen(name, name_size) != name_size - 1)
1065 if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
1066 !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
1069 *secctx_name = name;
1070 *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
1071 *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
1076 static int mdt_file_encctx_unpack(struct req_capsule *pill,
1077 void **encctx, size_t *encctx_size)
1082 if (!exp_connect_encrypt(pill->rc_req->rq_export))
1085 if (!req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_CLIENT) ||
1086 !req_capsule_field_present(pill, &RMF_FILE_ENCCTX, RCL_CLIENT))
1089 *encctx_size = req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_CLIENT);
1090 if (*encctx_size == 0)
1093 *encctx = req_capsule_client_get(pill, &RMF_FILE_ENCCTX);
1098 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
1100 struct lu_ucred *uc = mdt_ucred(info);
1101 struct md_attr *ma = &info->mti_attr;
1102 struct lu_attr *la = &ma->ma_attr;
1103 struct req_capsule *pill = info->mti_pill;
1104 struct mdt_reint_record *rr = &info->mti_rr;
1105 struct mdt_rec_setattr *rec;
1106 struct lu_nodemap *nodemap;
1110 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1111 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1115 /* This prior initialization is needed for old_init_ucred_reint() */
1116 uc->uc_fsuid = rec->sa_fsuid;
1117 uc->uc_fsgid = rec->sa_fsgid;
1118 uc->uc_cap = rec->sa_cap;
1119 uc->uc_suppgids[0] = rec->sa_suppgid;
1120 uc->uc_suppgids[1] = -1;
1122 rr->rr_fid1 = &rec->sa_fid;
1123 la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
1124 la->la_mode = rec->sa_mode;
1125 la->la_flags = rec->sa_attr_flags;
1127 nodemap = nodemap_get_from_exp(info->mti_exp);
1128 if (IS_ERR(nodemap))
1129 RETURN(PTR_ERR(nodemap));
1131 la->la_uid = nodemap_map_id(nodemap, NODEMAP_UID,
1132 NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1133 la->la_gid = nodemap_map_id(nodemap, NODEMAP_GID,
1134 NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1135 la->la_projid = rec->sa_projid;
1136 nodemap_putref(nodemap);
1138 la->la_size = rec->sa_size;
1139 la->la_blocks = rec->sa_blocks;
1140 la->la_ctime = rec->sa_ctime;
1141 la->la_atime = rec->sa_atime;
1142 la->la_mtime = rec->sa_mtime;
1143 ma->ma_valid = MA_INODE;
1145 ma->ma_attr_flags |= rec->sa_bias & (MDS_CLOSE_INTENT |
1146 MDS_DATA_MODIFIED | MDS_TRUNC_KEEP_LEASE |
1151 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
1153 struct req_capsule *pill = info->mti_pill;
1154 struct mdt_ioepoch *ioepoch;
1157 if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1158 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1162 if (ioepoch == NULL)
1165 info->mti_open_handle = ioepoch->mio_open_handle;
1170 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1171 struct req_capsule *pill = info->mti_pill;
1173 if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1174 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1175 if (info->mti_dlm_req == NULL)
1182 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1184 struct mdt_reint_record *rr = &info->mti_rr;
1185 struct md_attr *ma = &info->mti_attr;
1186 struct req_capsule *pill = info->mti_pill;
1190 rc = mdt_setattr_unpack_rec(info);
1194 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1195 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1196 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1199 if (rr->rr_eadatalen > 0) {
1200 const struct lmv_user_md *lum;
1202 lum = rr->rr_eadata;
1203 /* Sigh ma_valid(from req) does not indicate whether
1204 * it will set LOV/LMV EA, so we have to check magic */
1205 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1206 ma->ma_valid |= MA_LMV;
1207 ma->ma_lmv = (void *)rr->rr_eadata;
1208 ma->ma_lmv_size = rr->rr_eadatalen;
1210 ma->ma_valid |= MA_LOV;
1211 ma->ma_lmm = (void *)rr->rr_eadata;
1212 ma->ma_lmm_size = rr->rr_eadatalen;
1217 rc = mdt_dlmreq_unpack(info);
1221 static int mdt_close_intent_unpack(struct mdt_thread_info *info)
1223 struct md_attr *ma = &info->mti_attr;
1224 struct req_capsule *pill = info->mti_pill;
1227 if (!(ma->ma_attr_flags & MDS_CLOSE_INTENT))
1230 req_capsule_extend(pill, &RQF_MDS_CLOSE_INTENT);
1232 if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1233 req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1239 int mdt_close_unpack(struct mdt_thread_info *info)
1244 rc = mdt_close_handle_unpack(info);
1248 rc = mdt_setattr_unpack_rec(info);
1252 rc = mdt_close_intent_unpack(info);
1256 RETURN(mdt_init_ucred_reint(info));
1259 static int mdt_create_unpack(struct mdt_thread_info *info)
1261 struct lu_ucred *uc = mdt_ucred(info);
1262 struct mdt_rec_create *rec;
1263 struct lu_attr *attr = &info->mti_attr.ma_attr;
1264 struct mdt_reint_record *rr = &info->mti_rr;
1265 struct req_capsule *pill = info->mti_pill;
1266 struct md_op_spec *sp = &info->mti_spec;
1271 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1272 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1276 /* This prior initialization is needed for old_init_ucred_reint() */
1277 uc->uc_fsuid = rec->cr_fsuid;
1278 uc->uc_fsgid = rec->cr_fsgid;
1279 uc->uc_cap = rec->cr_cap;
1280 uc->uc_suppgids[0] = rec->cr_suppgid1;
1281 uc->uc_suppgids[1] = -1;
1282 uc->uc_umask = rec->cr_umask;
1284 rr->rr_fid1 = &rec->cr_fid1;
1285 rr->rr_fid2 = &rec->cr_fid2;
1286 attr->la_mode = rec->cr_mode;
1287 attr->la_rdev = rec->cr_rdev;
1288 attr->la_uid = rec->cr_fsuid;
1289 attr->la_gid = rec->cr_fsgid;
1290 attr->la_ctime = rec->cr_time;
1291 attr->la_mtime = rec->cr_time;
1292 attr->la_atime = rec->cr_time;
1293 attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1294 LA_CTIME | LA_MTIME | LA_ATIME;
1295 memset(&sp->u, 0, sizeof(sp->u));
1296 sp->sp_cr_flags = get_mrc_cr_flags(rec);
1298 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1302 if (S_ISLNK(attr->la_mode)) {
1303 const char *tgt = NULL;
1305 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1306 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1307 tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1308 sp->u.sp_symname = tgt;
1313 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1314 if (S_ISDIR(attr->la_mode) &&
1315 req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1316 sp->u.sp_ea.eadata =
1317 req_capsule_client_get(pill, &RMF_EADATA);
1318 sp->u.sp_ea.eadatalen =
1319 req_capsule_get_size(pill, &RMF_EADATA,
1321 sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1325 rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1326 &sp->sp_cr_file_secctx,
1327 &sp->sp_cr_file_secctx_size);
1331 rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
1332 &sp->sp_cr_file_encctx_size);
1336 rc = req_check_sepol(pill);
1340 rc = mdt_dlmreq_unpack(info);
1344 static int mdt_link_unpack(struct mdt_thread_info *info)
1346 struct lu_ucred *uc = mdt_ucred(info);
1347 struct mdt_rec_link *rec;
1348 struct lu_attr *attr = &info->mti_attr.ma_attr;
1349 struct mdt_reint_record *rr = &info->mti_rr;
1350 struct req_capsule *pill = info->mti_pill;
1355 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1356 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1360 /* This prior initialization is needed for old_init_ucred_reint() */
1361 uc->uc_fsuid = rec->lk_fsuid;
1362 uc->uc_fsgid = rec->lk_fsgid;
1363 uc->uc_cap = rec->lk_cap;
1364 uc->uc_suppgids[0] = rec->lk_suppgid1;
1365 uc->uc_suppgids[1] = rec->lk_suppgid2;
1367 attr->la_uid = rec->lk_fsuid;
1368 attr->la_gid = rec->lk_fsgid;
1369 rr->rr_fid1 = &rec->lk_fid1;
1370 rr->rr_fid2 = &rec->lk_fid2;
1371 attr->la_ctime = rec->lk_time;
1372 attr->la_mtime = rec->lk_time;
1373 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1375 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1379 rc = req_check_sepol(pill);
1383 rc = mdt_dlmreq_unpack(info);
1388 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1390 struct lu_ucred *uc = mdt_ucred(info);
1391 struct mdt_rec_unlink *rec;
1392 struct lu_attr *attr = &info->mti_attr.ma_attr;
1393 struct mdt_reint_record *rr = &info->mti_rr;
1394 struct req_capsule *pill = info->mti_pill;
1399 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1400 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1404 /* This prior initialization is needed for old_init_ucred_reint() */
1405 uc->uc_fsuid = rec->ul_fsuid;
1406 uc->uc_fsgid = rec->ul_fsgid;
1407 uc->uc_cap = rec->ul_cap;
1408 uc->uc_suppgids[0] = rec->ul_suppgid1;
1409 uc->uc_suppgids[1] = -1;
1411 attr->la_uid = rec->ul_fsuid;
1412 attr->la_gid = rec->ul_fsgid;
1413 rr->rr_fid1 = &rec->ul_fid1;
1414 rr->rr_fid2 = &rec->ul_fid2;
1415 attr->la_ctime = rec->ul_time;
1416 attr->la_mtime = rec->ul_time;
1417 attr->la_mode = rec->ul_mode;
1418 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1420 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1424 info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1426 rc = req_check_sepol(pill);
1430 rc = mdt_dlmreq_unpack(info);
1434 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1436 info->mti_spec.sp_rm_entry = 1;
1437 return mdt_unlink_unpack(info);
1440 static int mdt_rename_unpack(struct mdt_thread_info *info)
1442 struct lu_ucred *uc = mdt_ucred(info);
1443 struct mdt_rec_rename *rec;
1444 struct lu_attr *attr = &info->mti_attr.ma_attr;
1445 struct mdt_reint_record *rr = &info->mti_rr;
1446 struct req_capsule *pill = info->mti_pill;
1447 struct md_op_spec *spec = &info->mti_spec;
1452 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1453 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1457 /* This prior initialization is needed for old_init_ucred_reint() */
1458 uc->uc_fsuid = rec->rn_fsuid;
1459 uc->uc_fsgid = rec->rn_fsgid;
1460 uc->uc_cap = rec->rn_cap;
1461 uc->uc_suppgids[0] = rec->rn_suppgid1;
1462 uc->uc_suppgids[1] = rec->rn_suppgid2;
1464 attr->la_uid = rec->rn_fsuid;
1465 attr->la_gid = rec->rn_fsgid;
1466 rr->rr_fid1 = &rec->rn_fid1;
1467 rr->rr_fid2 = &rec->rn_fid2;
1468 attr->la_ctime = rec->rn_time;
1469 attr->la_mtime = rec->rn_time;
1470 /* rename_tgt contains the mode already */
1471 attr->la_mode = rec->rn_mode;
1472 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1474 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1478 rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1482 spec->no_create = !!req_is_replay(mdt_info_req(info));
1484 rc = req_check_sepol(pill);
1488 rc = mdt_dlmreq_unpack(info);
1493 static int mdt_migrate_unpack(struct mdt_thread_info *info)
1495 struct lu_ucred *uc = mdt_ucred(info);
1496 struct mdt_rec_rename *rec;
1497 struct lu_attr *attr = &info->mti_attr.ma_attr;
1498 struct mdt_reint_record *rr = &info->mti_rr;
1499 struct req_capsule *pill = info->mti_pill;
1500 struct md_op_spec *spec = &info->mti_spec;
1505 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1506 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1510 /* This prior initialization is needed for old_init_ucred_reint() */
1511 uc->uc_fsuid = rec->rn_fsuid;
1512 uc->uc_fsgid = rec->rn_fsgid;
1513 uc->uc_cap = rec->rn_cap;
1514 uc->uc_suppgids[0] = rec->rn_suppgid1;
1515 uc->uc_suppgids[1] = rec->rn_suppgid2;
1517 attr->la_uid = rec->rn_fsuid;
1518 attr->la_gid = rec->rn_fsgid;
1519 rr->rr_fid1 = &rec->rn_fid1;
1520 rr->rr_fid2 = &rec->rn_fid2;
1521 attr->la_ctime = rec->rn_time;
1522 attr->la_mtime = rec->rn_time;
1523 /* rename_tgt contains the mode already */
1524 attr->la_mode = rec->rn_mode;
1525 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1526 spec->sp_cr_flags = 0;
1528 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1532 if (rec->rn_bias & MDS_CLOSE_MIGRATE) {
1533 rc = mdt_close_handle_unpack(info);
1537 spec->sp_migrate_close = 1;
1539 spec->sp_migrate_close = 0;
1541 spec->sp_migrate_nsonly = 0;
1543 /* lustre version > 2.11 migration packs lum */
1544 if (req_capsule_has_field(pill, &RMF_EADATA, RCL_CLIENT)) {
1545 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1546 rr->rr_eadatalen = req_capsule_get_size(pill,
1550 if (rr->rr_eadatalen > 0) {
1551 rr->rr_eadata = req_capsule_client_get(pill,
1553 spec->u.sp_ea.eadatalen = rr->rr_eadatalen;
1554 spec->u.sp_ea.eadata = rr->rr_eadata;
1555 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1558 /* old client doesn't provide lum. */
1559 RETURN(-EOPNOTSUPP);
1563 spec->no_create = !!req_is_replay(mdt_info_req(info));
1565 rc = mdt_dlmreq_unpack(info);
1571 * please see comment above LOV_MAGIC_V1_DEFINED
1573 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata)
1575 struct lov_user_md_v1 *v1 = eadata;
1579 if (unlikely(req_is_replay(mdt_info_req(info)))) {
1580 if ((v1->lmm_magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
1581 v1->lmm_magic |= LOV_MAGIC_DEFINED;
1582 else if ((v1->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
1583 __swab32(LOV_MAGIC_MAGIC))
1584 v1->lmm_magic |= __swab32(LOV_MAGIC_DEFINED);
1588 static int mdt_open_unpack(struct mdt_thread_info *info)
1590 struct lu_ucred *uc = mdt_ucred(info);
1591 struct mdt_rec_create *rec;
1592 struct lu_attr *attr = &info->mti_attr.ma_attr;
1593 struct req_capsule *pill = info->mti_pill;
1594 struct mdt_reint_record *rr = &info->mti_rr;
1595 struct ptlrpc_request *req = mdt_info_req(info);
1596 struct md_op_spec *sp = &info->mti_spec;
1600 BUILD_BUG_ON(sizeof(struct mdt_rec_create) !=
1601 sizeof(struct mdt_rec_reint));
1602 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1606 /* This prior initialization is needed for old_init_ucred_reint() */
1607 uc->uc_fsuid = rec->cr_fsuid;
1608 uc->uc_fsgid = rec->cr_fsgid;
1609 uc->uc_cap = rec->cr_cap;
1610 uc->uc_suppgids[0] = rec->cr_suppgid1;
1611 uc->uc_suppgids[1] = rec->cr_suppgid2;
1612 uc->uc_umask = rec->cr_umask;
1614 rr->rr_fid1 = &rec->cr_fid1;
1615 rr->rr_fid2 = &rec->cr_fid2;
1616 rr->rr_open_handle = &rec->cr_open_handle_old;
1617 attr->la_mode = rec->cr_mode;
1618 attr->la_rdev = rec->cr_rdev;
1619 attr->la_uid = rec->cr_fsuid;
1620 attr->la_gid = rec->cr_fsgid;
1621 attr->la_ctime = rec->cr_time;
1622 attr->la_mtime = rec->cr_time;
1623 attr->la_atime = rec->cr_time;
1624 attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID |
1625 LA_CTIME | LA_MTIME | LA_ATIME;
1626 memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1627 info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1628 /* Do not trigger ASSERTION if client miss to set such flags. */
1629 if (unlikely(info->mti_spec.sp_cr_flags == 0))
1632 info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1634 mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1636 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1637 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1640 if (rr->rr_eadatalen > 0) {
1641 rr->rr_eadata = req_capsule_client_get(pill,
1643 sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1644 sp->u.sp_ea.eadata = rr->rr_eadata;
1645 sp->sp_archive_id = rec->cr_archive_id;
1646 sp->no_create = !!req_is_replay(req);
1647 mdt_fix_lov_magic(info, rr->rr_eadata);
1651 * Client default md_size may be 0 right after client start,
1652 * until all osc are connected, set here just some reasonable
1653 * value to prevent misbehavior.
1655 if (rr->rr_eadatalen == 0 &&
1656 !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1657 rr->rr_eadatalen = MIN_MD_SIZE;
1660 rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1661 &sp->sp_cr_file_secctx,
1662 &sp->sp_cr_file_secctx_size);
1666 rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
1667 &sp->sp_cr_file_encctx_size);
1671 rc = req_check_sepol(pill);
1678 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1680 struct mdt_reint_record *rr = &info->mti_rr;
1681 struct lu_ucred *uc = mdt_ucred(info);
1682 struct lu_attr *attr = &info->mti_attr.ma_attr;
1683 struct req_capsule *pill = info->mti_pill;
1684 struct mdt_rec_setxattr *rec;
1689 BUILD_BUG_ON(sizeof(struct mdt_rec_setxattr) !=
1690 sizeof(struct mdt_rec_reint));
1691 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1695 /* This prior initialization is needed for old_init_ucred_reint() */
1696 uc->uc_fsuid = rec->sx_fsuid;
1697 uc->uc_fsgid = rec->sx_fsgid;
1698 uc->uc_cap = rec->sx_cap;
1699 uc->uc_suppgids[0] = rec->sx_suppgid1;
1700 uc->uc_suppgids[1] = -1;
1702 rr->rr_opcode = rec->sx_opcode;
1703 rr->rr_fid1 = &rec->sx_fid;
1704 attr->la_valid = rec->sx_valid;
1705 attr->la_ctime = rec->sx_time;
1706 attr->la_size = rec->sx_size;
1707 attr->la_flags = rec->sx_flags;
1709 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1713 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1714 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1717 if (rr->rr_eadatalen > info->mti_mdt->mdt_max_ea_size)
1720 if (rr->rr_eadatalen > 0) {
1721 rr->rr_eadata = req_capsule_client_get(pill,
1723 if (rr->rr_eadata == NULL)
1726 rr->rr_eadata = NULL;
1728 } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1729 CDEBUG(D_INFO, "no xattr data supplied\n");
1733 rc = req_check_sepol(pill);
1737 if (mdt_dlmreq_unpack(info) < 0)
1743 static int mdt_resync_unpack(struct mdt_thread_info *info)
1745 struct req_capsule *pill = info->mti_pill;
1746 struct mdt_reint_record *rr = &info->mti_rr;
1747 struct lu_ucred *uc = mdt_ucred(info);
1748 struct mdt_rec_resync *rec;
1751 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1752 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1756 /* This prior initialization is needed for old_init_ucred_reint() */
1757 uc->uc_fsuid = rec->rs_fsuid;
1758 uc->uc_fsgid = rec->rs_fsgid;
1759 uc->uc_cap = rec->rs_cap;
1761 rr->rr_fid1 = &rec->rs_fid;
1762 rr->rr_mirror_id = rec->rs_mirror_id;
1764 /* cookie doesn't need to be swapped but it has been swapped
1765 * in lustre_swab_mdt_rec_reint() as rr_mtime, so here it needs
1767 if (ptlrpc_req_need_swab(mdt_info_req(info)))
1768 __swab64s(&rec->rs_lease_handle.cookie);
1769 rr->rr_lease_handle = &rec->rs_lease_handle;
1771 RETURN(mdt_dlmreq_unpack(info));
1774 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1776 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1777 [REINT_SETATTR] = mdt_setattr_unpack,
1778 [REINT_CREATE] = mdt_create_unpack,
1779 [REINT_LINK] = mdt_link_unpack,
1780 [REINT_UNLINK] = mdt_unlink_unpack,
1781 [REINT_RENAME] = mdt_rename_unpack,
1782 [REINT_OPEN] = mdt_open_unpack,
1783 [REINT_SETXATTR] = mdt_setxattr_unpack,
1784 [REINT_RMENTRY] = mdt_rmentry_unpack,
1785 [REINT_MIGRATE] = mdt_migrate_unpack,
1786 [REINT_RESYNC] = mdt_resync_unpack,
1789 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1794 memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1795 if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1796 info->mti_rr.rr_opcode = op;
1797 rc = mdt_reint_unpackers[op](info);
1799 CERROR("Unexpected opcode %d\n", op);
1805 int mdt_pack_secctx_in_reply(struct mdt_thread_info *info,
1806 struct mdt_object *child)
1809 struct lu_buf *buffer;
1810 struct mdt_body *repbody;
1811 struct req_capsule *pill = info->mti_pill;
1814 if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
1815 req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0) {
1817 req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1818 buffer = &info->mti_buf;
1820 /* fill reply buffer with security context now */
1821 buffer->lb_len = req_capsule_get_size(pill, &RMF_FILE_SECCTX,
1823 buffer->lb_buf = req_capsule_server_get(info->mti_pill,
1825 rc = mo_xattr_get(info->mti_env, mdt_object_child(child),
1826 buffer, secctx_name);
1829 "found security context of size %d for "DFID"\n",
1830 rc, PFID(mdt_object_fid(child)));
1832 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1833 repbody->mbo_valid |= OBD_MD_SECCTX;
1834 if (rc < buffer->lb_len)
1835 req_capsule_shrink(pill, &RMF_FILE_SECCTX, rc,
1840 "security context not found for "DFID": rc = %d\n",
1841 PFID(mdt_object_fid(child)), rc);
1842 req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0,
1844 /* handling -ENOENT is important because it may change
1845 * object state in DNE env dropping LOHA_EXISTS flag,
1846 * it is important to return that to the caller.
1847 * Check LU-13115 for details.
1856 /* check whether two FIDs belong to different MDT. */
1857 static int mdt_fids_different_target(struct mdt_thread_info *info,
1858 const struct lu_fid *fid1,
1859 const struct lu_fid *fid2)
1861 const struct lu_env *env = info->mti_env;
1862 struct mdt_device *mdt = info->mti_mdt;
1863 struct lu_seq_range *range = &info->mti_range;
1864 struct seq_server_site *ss;
1865 __u32 index1, index2;
1868 if (fid_seq(fid1) == fid_seq(fid2))
1871 ss = mdt->mdt_lu_dev.ld_site->ld_seq_site;
1873 range->lsr_flags = LU_SEQ_RANGE_MDT;
1874 rc = fld_server_lookup(env, ss->ss_server_fld, fid1->f_seq, range);
1878 index1 = range->lsr_index;
1880 rc = fld_server_lookup(env, ss->ss_server_fld, fid2->f_seq, range);
1884 index2 = range->lsr_index;
1886 return index1 != index2;
1890 * Check whether \a child is remote object on \a parent.
1892 * \param[in] info thread environment
1893 * \param[in] parent parent object, it's the same as child object in
1895 * \param[in] child child object
1897 * \retval 1 is remote object.
1898 * \retval 0 isn't remote object.
1899 * \retval < 1 error code
1901 int mdt_is_remote_object(struct mdt_thread_info *info,
1902 struct mdt_object *parent,
1903 struct mdt_object *child)
1905 struct lu_buf *buf = &info->mti_big_buf;
1906 struct linkea_data ldata = { NULL };
1907 struct link_ea_header *leh;
1908 struct link_ea_entry *lee;
1909 struct lu_name name;
1917 if (fid_is_root(mdt_object_fid(child)))
1920 if (likely(parent != child)) {
1921 if (mdt_object_remote(parent) ^ mdt_object_remote(child))
1924 if (!mdt_object_remote(parent) && !mdt_object_remote(child))
1927 rc = mdt_fids_different_target(info, mdt_object_fid(parent),
1928 mdt_object_fid(child));
1932 /* client < 2.13.52 getattr_by_fid parent and child are the same */
1933 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
1938 rc = mdt_links_read(info, child, &ldata);
1939 /* can't read linkea, just assume it's remote object */
1940 if (rc == -ENOENT || rc == -ENODATA)
1946 lee = (struct link_ea_entry *)(leh + 1);
1947 for (i = 0; i < leh->leh_reccount; i++) {
1948 linkea_entry_unpack(lee, &reclen, &name, &pfid);
1949 lee = (struct link_ea_entry *) ((char *)lee + reclen);
1950 if (mdt_fids_different_target(info, &pfid,
1951 mdt_object_fid(child)))
1958 int mdt_pack_encctx_in_reply(struct mdt_thread_info *info,
1959 struct mdt_object *child)
1961 struct lu_buf *buffer;
1962 struct mdt_body *repbody;
1963 struct req_capsule *pill = info->mti_pill;
1964 struct obd_export *exp = mdt_info_req(info)->rq_export;
1967 if (!exp_connect_encrypt(exp))
1970 if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
1971 req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0) {
1972 struct lu_attr la = { 0 };
1973 struct dt_object *dt = mdt_obj2dt(child);
1975 if (dt && dt->do_ops && dt->do_ops->do_attr_get)
1976 dt_attr_get(info->mti_env, mdt_obj2dt(child), &la);
1978 if (la.la_valid & LA_FLAGS && la.la_flags & LUSTRE_ENCRYPT_FL) {
1979 buffer = &info->mti_buf;
1981 /* fill reply buffer with encryption context now */
1983 req_capsule_get_size(pill, &RMF_FILE_ENCCTX,
1986 req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
1987 rc = mo_xattr_get(info->mti_env,
1988 mdt_object_child(child),
1990 LL_XATTR_NAME_ENCRYPTION_CONTEXT);
1993 "found encryption ctx of size %d for "DFID"\n",
1994 rc, PFID(mdt_object_fid(child)));
1996 repbody = req_capsule_server_get(pill,
1998 repbody->mbo_valid |= OBD_MD_ENCCTX;
1999 if (rc < buffer->lb_len)
2000 req_capsule_shrink(pill,
2001 &RMF_FILE_ENCCTX, rc,
2006 "encryption ctx not found for "DFID": rc = %d\n",
2007 PFID(mdt_object_fid(child)), rc);
2008 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,
2010 /* handling -ENOENT is important because it may
2011 * change object state in DNE env dropping
2012 * LOHA_EXISTS flag, it is important to return
2013 * that to the caller.
2014 * Check LU-13115 for details.
2020 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,