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(__u64 in, struct mdt_reint_record *rr,
959 if (in & MDS_ATTR_MODE)
961 if (in & MDS_ATTR_UID)
963 if (in & MDS_ATTR_GID)
965 if (in & MDS_ATTR_SIZE)
967 if (in & MDS_ATTR_BLOCKS)
969 if (in & MDS_ATTR_ATIME_SET)
971 if (in & MDS_ATTR_CTIME_SET)
973 if (in & MDS_ATTR_MTIME_SET)
975 if (in & MDS_ATTR_ATTR_FLAG)
977 if (in & MDS_ATTR_KILL_SUID)
979 if (in & MDS_ATTR_KILL_SGID)
981 if (in & MDS_ATTR_PROJID)
983 if (in & MDS_ATTR_LSIZE)
985 if (in & MDS_ATTR_LBLOCKS)
988 if (in & MDS_ATTR_FROM_OPEN)
989 rr->rr_flags |= MRF_OPEN_TRUNC;
990 if (in & MDS_ATTR_OVERRIDE)
991 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
992 if (in & MDS_ATTR_FORCE)
993 ma->ma_attr_flags |= MDS_PERM_BYPASS;
995 in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | MDS_ATTR_PROJID |
996 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
997 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
998 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
999 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
1000 MDS_ATTR_FROM_OPEN | MDS_ATTR_LSIZE | MDS_ATTR_LBLOCKS |
1003 CDEBUG(D_INFO, "Unknown attr bits: %#llx\n", in);
1009 int mdt_name_unpack(struct req_capsule *pill,
1010 const struct req_msg_field *field,
1012 enum mdt_name_flags flags)
1014 ln->ln_name = req_capsule_client_get(pill, field);
1015 ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
1017 if (!lu_name_is_valid(ln)) {
1024 if ((flags & MNF_FIX_ANON) &&
1025 ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
1026 /* Newer (3.x) kernels use a name of "/" for the
1027 * "anonymous" disconnected dentries from NFS
1028 * filehandle conversion. See d_obtain_alias(). */
1036 static int mdt_file_secctx_unpack(struct req_capsule *pill,
1037 const char **secctx_name,
1038 void **secctx, size_t *secctx_size)
1043 *secctx_name = NULL;
1047 if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
1048 !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
1051 name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
1056 if (name_size > XATTR_NAME_MAX + 1)
1059 name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1060 if (strnlen(name, name_size) != name_size - 1)
1063 if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
1064 !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
1067 *secctx_name = name;
1068 *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
1069 *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
1074 static int mdt_file_encctx_unpack(struct req_capsule *pill,
1075 void **encctx, size_t *encctx_size)
1080 if (!exp_connect_encrypt(pill->rc_req->rq_export))
1083 if (!req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_CLIENT) ||
1084 !req_capsule_field_present(pill, &RMF_FILE_ENCCTX, RCL_CLIENT))
1087 *encctx_size = req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_CLIENT);
1088 if (*encctx_size == 0)
1091 *encctx = req_capsule_client_get(pill, &RMF_FILE_ENCCTX);
1096 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
1098 struct lu_ucred *uc = mdt_ucred(info);
1099 struct md_attr *ma = &info->mti_attr;
1100 struct lu_attr *la = &ma->ma_attr;
1101 struct req_capsule *pill = info->mti_pill;
1102 struct mdt_reint_record *rr = &info->mti_rr;
1103 struct mdt_rec_setattr *rec;
1104 struct lu_nodemap *nodemap;
1108 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1109 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1113 /* This prior initialization is needed for old_init_ucred_reint() */
1114 uc->uc_fsuid = rec->sa_fsuid;
1115 uc->uc_fsgid = rec->sa_fsgid;
1116 uc->uc_cap = rec->sa_cap;
1117 uc->uc_suppgids[0] = rec->sa_suppgid;
1118 uc->uc_suppgids[1] = -1;
1120 rr->rr_fid1 = &rec->sa_fid;
1121 la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
1122 la->la_mode = rec->sa_mode;
1123 la->la_flags = rec->sa_attr_flags;
1125 nodemap = nodemap_get_from_exp(info->mti_exp);
1126 if (IS_ERR(nodemap))
1127 RETURN(PTR_ERR(nodemap));
1129 la->la_uid = nodemap_map_id(nodemap, NODEMAP_UID,
1130 NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1131 la->la_gid = nodemap_map_id(nodemap, NODEMAP_GID,
1132 NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1133 la->la_projid = rec->sa_projid;
1134 nodemap_putref(nodemap);
1136 la->la_size = rec->sa_size;
1137 la->la_blocks = rec->sa_blocks;
1138 la->la_ctime = rec->sa_ctime;
1139 la->la_atime = rec->sa_atime;
1140 la->la_mtime = rec->sa_mtime;
1141 ma->ma_valid = MA_INODE;
1143 ma->ma_attr_flags |= rec->sa_bias & (MDS_CLOSE_INTENT |
1144 MDS_DATA_MODIFIED | MDS_TRUNC_KEEP_LEASE |
1149 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
1151 struct req_capsule *pill = info->mti_pill;
1152 struct mdt_ioepoch *ioepoch;
1155 if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1156 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1160 if (ioepoch == NULL)
1163 info->mti_open_handle = ioepoch->mio_open_handle;
1168 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1169 struct req_capsule *pill = info->mti_pill;
1171 if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1172 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1173 if (info->mti_dlm_req == NULL)
1180 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1182 struct mdt_reint_record *rr = &info->mti_rr;
1183 struct md_attr *ma = &info->mti_attr;
1184 struct req_capsule *pill = info->mti_pill;
1188 rc = mdt_setattr_unpack_rec(info);
1192 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1193 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1194 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1197 if (rr->rr_eadatalen > 0) {
1198 const struct lmv_user_md *lum;
1200 lum = rr->rr_eadata;
1201 /* Sigh ma_valid(from req) does not indicate whether
1202 * it will set LOV/LMV EA, so we have to check magic */
1203 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1204 ma->ma_valid |= MA_LMV;
1205 ma->ma_lmv = (void *)rr->rr_eadata;
1206 ma->ma_lmv_size = rr->rr_eadatalen;
1208 ma->ma_valid |= MA_LOV;
1209 ma->ma_lmm = (void *)rr->rr_eadata;
1210 ma->ma_lmm_size = rr->rr_eadatalen;
1215 rc = mdt_dlmreq_unpack(info);
1219 static int mdt_close_intent_unpack(struct mdt_thread_info *info)
1221 struct md_attr *ma = &info->mti_attr;
1222 struct req_capsule *pill = info->mti_pill;
1225 if (!(ma->ma_attr_flags & MDS_CLOSE_INTENT))
1228 req_capsule_extend(pill, &RQF_MDS_CLOSE_INTENT);
1230 if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1231 req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1237 int mdt_close_unpack(struct mdt_thread_info *info)
1242 rc = mdt_close_handle_unpack(info);
1246 rc = mdt_setattr_unpack_rec(info);
1250 rc = mdt_close_intent_unpack(info);
1254 RETURN(mdt_init_ucred_reint(info));
1257 static int mdt_create_unpack(struct mdt_thread_info *info)
1259 struct lu_ucred *uc = mdt_ucred(info);
1260 struct mdt_rec_create *rec;
1261 struct lu_attr *attr = &info->mti_attr.ma_attr;
1262 struct mdt_reint_record *rr = &info->mti_rr;
1263 struct req_capsule *pill = info->mti_pill;
1264 struct md_op_spec *sp = &info->mti_spec;
1269 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1270 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1274 /* This prior initialization is needed for old_init_ucred_reint() */
1275 uc->uc_fsuid = rec->cr_fsuid;
1276 uc->uc_fsgid = rec->cr_fsgid;
1277 uc->uc_cap = rec->cr_cap;
1278 uc->uc_suppgids[0] = rec->cr_suppgid1;
1279 uc->uc_suppgids[1] = -1;
1280 uc->uc_umask = rec->cr_umask;
1282 rr->rr_fid1 = &rec->cr_fid1;
1283 rr->rr_fid2 = &rec->cr_fid2;
1284 attr->la_mode = rec->cr_mode;
1285 attr->la_rdev = rec->cr_rdev;
1286 attr->la_uid = rec->cr_fsuid;
1287 attr->la_gid = rec->cr_fsgid;
1288 attr->la_ctime = rec->cr_time;
1289 attr->la_mtime = rec->cr_time;
1290 attr->la_atime = rec->cr_time;
1291 attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1292 LA_CTIME | LA_MTIME | LA_ATIME;
1293 memset(&sp->u, 0, sizeof(sp->u));
1294 sp->sp_cr_flags = get_mrc_cr_flags(rec);
1296 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1300 if (S_ISLNK(attr->la_mode)) {
1301 const char *tgt = NULL;
1303 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1304 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1305 tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1306 sp->u.sp_symname = tgt;
1311 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1312 if (S_ISDIR(attr->la_mode) &&
1313 req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1314 sp->u.sp_ea.eadata =
1315 req_capsule_client_get(pill, &RMF_EADATA);
1316 sp->u.sp_ea.eadatalen =
1317 req_capsule_get_size(pill, &RMF_EADATA,
1319 sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1323 rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1324 &sp->sp_cr_file_secctx,
1325 &sp->sp_cr_file_secctx_size);
1329 rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
1330 &sp->sp_cr_file_encctx_size);
1334 rc = req_check_sepol(pill);
1338 rc = mdt_dlmreq_unpack(info);
1342 static int mdt_link_unpack(struct mdt_thread_info *info)
1344 struct lu_ucred *uc = mdt_ucred(info);
1345 struct mdt_rec_link *rec;
1346 struct lu_attr *attr = &info->mti_attr.ma_attr;
1347 struct mdt_reint_record *rr = &info->mti_rr;
1348 struct req_capsule *pill = info->mti_pill;
1353 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1354 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1358 /* This prior initialization is needed for old_init_ucred_reint() */
1359 uc->uc_fsuid = rec->lk_fsuid;
1360 uc->uc_fsgid = rec->lk_fsgid;
1361 uc->uc_cap = rec->lk_cap;
1362 uc->uc_suppgids[0] = rec->lk_suppgid1;
1363 uc->uc_suppgids[1] = rec->lk_suppgid2;
1365 attr->la_uid = rec->lk_fsuid;
1366 attr->la_gid = rec->lk_fsgid;
1367 rr->rr_fid1 = &rec->lk_fid1;
1368 rr->rr_fid2 = &rec->lk_fid2;
1369 attr->la_ctime = rec->lk_time;
1370 attr->la_mtime = rec->lk_time;
1371 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1373 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1377 rc = req_check_sepol(pill);
1381 rc = mdt_dlmreq_unpack(info);
1386 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1388 struct lu_ucred *uc = mdt_ucred(info);
1389 struct mdt_rec_unlink *rec;
1390 struct lu_attr *attr = &info->mti_attr.ma_attr;
1391 struct mdt_reint_record *rr = &info->mti_rr;
1392 struct req_capsule *pill = info->mti_pill;
1397 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1398 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1402 /* This prior initialization is needed for old_init_ucred_reint() */
1403 uc->uc_fsuid = rec->ul_fsuid;
1404 uc->uc_fsgid = rec->ul_fsgid;
1405 uc->uc_cap = rec->ul_cap;
1406 uc->uc_suppgids[0] = rec->ul_suppgid1;
1407 uc->uc_suppgids[1] = -1;
1409 attr->la_uid = rec->ul_fsuid;
1410 attr->la_gid = rec->ul_fsgid;
1411 rr->rr_fid1 = &rec->ul_fid1;
1412 rr->rr_fid2 = &rec->ul_fid2;
1413 attr->la_ctime = rec->ul_time;
1414 attr->la_mtime = rec->ul_time;
1415 attr->la_mode = rec->ul_mode;
1416 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1418 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1422 info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1424 rc = req_check_sepol(pill);
1428 rc = mdt_dlmreq_unpack(info);
1432 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1434 info->mti_spec.sp_rm_entry = 1;
1435 return mdt_unlink_unpack(info);
1438 static int mdt_rename_unpack(struct mdt_thread_info *info)
1440 struct lu_ucred *uc = mdt_ucred(info);
1441 struct mdt_rec_rename *rec;
1442 struct lu_attr *attr = &info->mti_attr.ma_attr;
1443 struct mdt_reint_record *rr = &info->mti_rr;
1444 struct req_capsule *pill = info->mti_pill;
1445 struct md_op_spec *spec = &info->mti_spec;
1450 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1451 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1455 /* This prior initialization is needed for old_init_ucred_reint() */
1456 uc->uc_fsuid = rec->rn_fsuid;
1457 uc->uc_fsgid = rec->rn_fsgid;
1458 uc->uc_cap = rec->rn_cap;
1459 uc->uc_suppgids[0] = rec->rn_suppgid1;
1460 uc->uc_suppgids[1] = rec->rn_suppgid2;
1462 attr->la_uid = rec->rn_fsuid;
1463 attr->la_gid = rec->rn_fsgid;
1464 rr->rr_fid1 = &rec->rn_fid1;
1465 rr->rr_fid2 = &rec->rn_fid2;
1466 attr->la_ctime = rec->rn_time;
1467 attr->la_mtime = rec->rn_time;
1468 /* rename_tgt contains the mode already */
1469 attr->la_mode = rec->rn_mode;
1470 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1472 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1476 rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1480 spec->no_create = !!req_is_replay(mdt_info_req(info));
1482 rc = req_check_sepol(pill);
1486 rc = mdt_dlmreq_unpack(info);
1491 static int mdt_migrate_unpack(struct mdt_thread_info *info)
1493 struct lu_ucred *uc = mdt_ucred(info);
1494 struct mdt_rec_rename *rec;
1495 struct lu_attr *attr = &info->mti_attr.ma_attr;
1496 struct mdt_reint_record *rr = &info->mti_rr;
1497 struct req_capsule *pill = info->mti_pill;
1498 struct md_op_spec *spec = &info->mti_spec;
1503 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1504 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1508 /* This prior initialization is needed for old_init_ucred_reint() */
1509 uc->uc_fsuid = rec->rn_fsuid;
1510 uc->uc_fsgid = rec->rn_fsgid;
1511 uc->uc_cap = rec->rn_cap;
1512 uc->uc_suppgids[0] = rec->rn_suppgid1;
1513 uc->uc_suppgids[1] = rec->rn_suppgid2;
1515 attr->la_uid = rec->rn_fsuid;
1516 attr->la_gid = rec->rn_fsgid;
1517 rr->rr_fid1 = &rec->rn_fid1;
1518 rr->rr_fid2 = &rec->rn_fid2;
1519 attr->la_ctime = rec->rn_time;
1520 attr->la_mtime = rec->rn_time;
1521 /* rename_tgt contains the mode already */
1522 attr->la_mode = rec->rn_mode;
1523 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1524 spec->sp_cr_flags = 0;
1526 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1530 if (rec->rn_bias & MDS_CLOSE_MIGRATE) {
1531 rc = mdt_close_handle_unpack(info);
1535 spec->sp_migrate_close = 1;
1537 spec->sp_migrate_close = 0;
1539 spec->sp_migrate_nsonly = 0;
1541 /* lustre version > 2.11 migration packs lum */
1542 if (req_capsule_has_field(pill, &RMF_EADATA, RCL_CLIENT)) {
1543 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1544 rr->rr_eadatalen = req_capsule_get_size(pill,
1548 if (rr->rr_eadatalen > 0) {
1549 rr->rr_eadata = req_capsule_client_get(pill,
1551 spec->u.sp_ea.eadatalen = rr->rr_eadatalen;
1552 spec->u.sp_ea.eadata = rr->rr_eadata;
1553 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1556 /* old client doesn't provide lum. */
1557 RETURN(-EOPNOTSUPP);
1561 spec->no_create = !!req_is_replay(mdt_info_req(info));
1563 rc = mdt_dlmreq_unpack(info);
1569 * please see comment above LOV_MAGIC_V1_DEFINED
1571 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata)
1573 struct lov_user_md_v1 *v1 = eadata;
1577 if (unlikely(req_is_replay(mdt_info_req(info)))) {
1578 if ((v1->lmm_magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
1579 v1->lmm_magic |= LOV_MAGIC_DEFINED;
1580 else if ((v1->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
1581 __swab32(LOV_MAGIC_MAGIC))
1582 v1->lmm_magic |= __swab32(LOV_MAGIC_DEFINED);
1586 static int mdt_open_unpack(struct mdt_thread_info *info)
1588 struct lu_ucred *uc = mdt_ucred(info);
1589 struct mdt_rec_create *rec;
1590 struct lu_attr *attr = &info->mti_attr.ma_attr;
1591 struct req_capsule *pill = info->mti_pill;
1592 struct mdt_reint_record *rr = &info->mti_rr;
1593 struct ptlrpc_request *req = mdt_info_req(info);
1594 struct md_op_spec *sp = &info->mti_spec;
1598 BUILD_BUG_ON(sizeof(struct mdt_rec_create) !=
1599 sizeof(struct mdt_rec_reint));
1600 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1604 /* This prior initialization is needed for old_init_ucred_reint() */
1605 uc->uc_fsuid = rec->cr_fsuid;
1606 uc->uc_fsgid = rec->cr_fsgid;
1607 uc->uc_cap = rec->cr_cap;
1608 uc->uc_suppgids[0] = rec->cr_suppgid1;
1609 uc->uc_suppgids[1] = rec->cr_suppgid2;
1610 uc->uc_umask = rec->cr_umask;
1612 rr->rr_fid1 = &rec->cr_fid1;
1613 rr->rr_fid2 = &rec->cr_fid2;
1614 rr->rr_open_handle = &rec->cr_open_handle_old;
1615 attr->la_mode = rec->cr_mode;
1616 attr->la_rdev = rec->cr_rdev;
1617 attr->la_uid = rec->cr_fsuid;
1618 attr->la_gid = rec->cr_fsgid;
1619 attr->la_ctime = rec->cr_time;
1620 attr->la_mtime = rec->cr_time;
1621 attr->la_atime = rec->cr_time;
1622 attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID |
1623 LA_CTIME | LA_MTIME | LA_ATIME;
1624 memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1625 info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1626 /* Do not trigger ASSERTION if client miss to set such flags. */
1627 if (unlikely(info->mti_spec.sp_cr_flags == 0))
1630 info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1632 mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1634 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1635 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1638 if (rr->rr_eadatalen > 0) {
1639 rr->rr_eadata = req_capsule_client_get(pill,
1641 sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1642 sp->u.sp_ea.eadata = rr->rr_eadata;
1643 sp->sp_archive_id = rec->cr_archive_id;
1644 sp->no_create = !!req_is_replay(req);
1645 mdt_fix_lov_magic(info, rr->rr_eadata);
1649 * Client default md_size may be 0 right after client start,
1650 * until all osc are connected, set here just some reasonable
1651 * value to prevent misbehavior.
1653 if (rr->rr_eadatalen == 0 &&
1654 !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1655 rr->rr_eadatalen = MIN_MD_SIZE;
1658 rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1659 &sp->sp_cr_file_secctx,
1660 &sp->sp_cr_file_secctx_size);
1664 rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
1665 &sp->sp_cr_file_encctx_size);
1669 rc = req_check_sepol(pill);
1676 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1678 struct mdt_reint_record *rr = &info->mti_rr;
1679 struct lu_ucred *uc = mdt_ucred(info);
1680 struct lu_attr *attr = &info->mti_attr.ma_attr;
1681 struct req_capsule *pill = info->mti_pill;
1682 struct mdt_rec_setxattr *rec;
1687 BUILD_BUG_ON(sizeof(struct mdt_rec_setxattr) !=
1688 sizeof(struct mdt_rec_reint));
1689 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1693 /* This prior initialization is needed for old_init_ucred_reint() */
1694 uc->uc_fsuid = rec->sx_fsuid;
1695 uc->uc_fsgid = rec->sx_fsgid;
1696 uc->uc_cap = rec->sx_cap;
1697 uc->uc_suppgids[0] = rec->sx_suppgid1;
1698 uc->uc_suppgids[1] = -1;
1700 rr->rr_opcode = rec->sx_opcode;
1701 rr->rr_fid1 = &rec->sx_fid;
1702 attr->la_valid = rec->sx_valid;
1703 attr->la_ctime = rec->sx_time;
1704 attr->la_size = rec->sx_size;
1705 attr->la_flags = rec->sx_flags;
1707 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1711 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1712 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1715 if (rr->rr_eadatalen > info->mti_mdt->mdt_max_ea_size)
1718 if (rr->rr_eadatalen > 0) {
1719 rr->rr_eadata = req_capsule_client_get(pill,
1721 if (rr->rr_eadata == NULL)
1724 rr->rr_eadata = NULL;
1726 } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1727 CDEBUG(D_INFO, "no xattr data supplied\n");
1731 rc = req_check_sepol(pill);
1735 if (mdt_dlmreq_unpack(info) < 0)
1741 static int mdt_resync_unpack(struct mdt_thread_info *info)
1743 struct req_capsule *pill = info->mti_pill;
1744 struct mdt_reint_record *rr = &info->mti_rr;
1745 struct lu_ucred *uc = mdt_ucred(info);
1746 struct mdt_rec_resync *rec;
1749 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1750 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1754 /* This prior initialization is needed for old_init_ucred_reint() */
1755 uc->uc_fsuid = rec->rs_fsuid;
1756 uc->uc_fsgid = rec->rs_fsgid;
1757 uc->uc_cap = rec->rs_cap;
1759 rr->rr_fid1 = &rec->rs_fid;
1760 rr->rr_mirror_id = rec->rs_mirror_id;
1762 /* cookie doesn't need to be swapped but it has been swapped
1763 * in lustre_swab_mdt_rec_reint() as rr_mtime, so here it needs
1765 if (ptlrpc_req_need_swab(mdt_info_req(info)))
1766 __swab64s(&rec->rs_lease_handle.cookie);
1767 rr->rr_lease_handle = &rec->rs_lease_handle;
1769 RETURN(mdt_dlmreq_unpack(info));
1772 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1774 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1775 [REINT_SETATTR] = mdt_setattr_unpack,
1776 [REINT_CREATE] = mdt_create_unpack,
1777 [REINT_LINK] = mdt_link_unpack,
1778 [REINT_UNLINK] = mdt_unlink_unpack,
1779 [REINT_RENAME] = mdt_rename_unpack,
1780 [REINT_OPEN] = mdt_open_unpack,
1781 [REINT_SETXATTR] = mdt_setxattr_unpack,
1782 [REINT_RMENTRY] = mdt_rmentry_unpack,
1783 [REINT_MIGRATE] = mdt_migrate_unpack,
1784 [REINT_RESYNC] = mdt_resync_unpack,
1787 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1792 memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1793 if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1794 info->mti_rr.rr_opcode = op;
1795 rc = mdt_reint_unpackers[op](info);
1797 CERROR("Unexpected opcode %d\n", op);
1803 int mdt_pack_secctx_in_reply(struct mdt_thread_info *info,
1804 struct mdt_object *child)
1807 struct lu_buf *buffer;
1808 struct mdt_body *repbody;
1809 struct req_capsule *pill = info->mti_pill;
1812 if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
1813 req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0) {
1815 req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1816 buffer = &info->mti_buf;
1818 /* fill reply buffer with security context now */
1819 buffer->lb_len = req_capsule_get_size(pill, &RMF_FILE_SECCTX,
1821 buffer->lb_buf = req_capsule_server_get(info->mti_pill,
1823 rc = mo_xattr_get(info->mti_env, mdt_object_child(child),
1824 buffer, secctx_name);
1827 "found security context of size %d for "DFID"\n",
1828 rc, PFID(mdt_object_fid(child)));
1830 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1831 repbody->mbo_valid |= OBD_MD_SECCTX;
1832 if (rc < buffer->lb_len)
1833 req_capsule_shrink(pill, &RMF_FILE_SECCTX, rc,
1838 "security context not found for "DFID": rc = %d\n",
1839 PFID(mdt_object_fid(child)), rc);
1840 req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0,
1842 /* handling -ENOENT is important because it may change
1843 * object state in DNE env dropping LOHA_EXISTS flag,
1844 * it is important to return that to the caller.
1845 * Check LU-13115 for details.
1854 /* check whether two FIDs belong to different MDT. */
1855 static int mdt_fids_different_target(struct mdt_thread_info *info,
1856 const struct lu_fid *fid1,
1857 const struct lu_fid *fid2)
1859 const struct lu_env *env = info->mti_env;
1860 struct mdt_device *mdt = info->mti_mdt;
1861 struct lu_seq_range *range = &info->mti_range;
1862 struct seq_server_site *ss;
1863 __u32 index1, index2;
1866 if (fid_seq(fid1) == fid_seq(fid2))
1869 ss = mdt->mdt_lu_dev.ld_site->ld_seq_site;
1871 range->lsr_flags = LU_SEQ_RANGE_MDT;
1872 rc = fld_server_lookup(env, ss->ss_server_fld, fid1->f_seq, range);
1876 index1 = range->lsr_index;
1878 rc = fld_server_lookup(env, ss->ss_server_fld, fid2->f_seq, range);
1882 index2 = range->lsr_index;
1884 return index1 != index2;
1888 * Check whether \a child is remote object on \a parent.
1890 * \param[in] info thread environment
1891 * \param[in] parent parent object, it's the same as child object in
1893 * \param[in] child child object
1895 * \retval 1 is remote object.
1896 * \retval 0 isn't remote object.
1897 * \retval < 1 error code
1899 int mdt_is_remote_object(struct mdt_thread_info *info,
1900 struct mdt_object *parent,
1901 struct mdt_object *child)
1903 struct lu_buf *buf = &info->mti_big_buf;
1904 struct linkea_data ldata = { NULL };
1905 struct link_ea_header *leh;
1906 struct link_ea_entry *lee;
1907 struct lu_name name;
1915 if (fid_is_root(mdt_object_fid(child)))
1918 if (likely(parent != child)) {
1919 if (mdt_object_remote(parent) ^ mdt_object_remote(child))
1922 if (!mdt_object_remote(parent) && !mdt_object_remote(child))
1925 rc = mdt_fids_different_target(info, mdt_object_fid(parent),
1926 mdt_object_fid(child));
1930 /* client < 2.13.52 getattr_by_fid parent and child are the same */
1931 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
1936 rc = mdt_links_read(info, child, &ldata);
1937 /* can't read linkea, just assume it's remote object */
1938 if (rc == -ENOENT || rc == -ENODATA)
1944 lee = (struct link_ea_entry *)(leh + 1);
1945 for (i = 0; i < leh->leh_reccount; i++) {
1946 linkea_entry_unpack(lee, &reclen, &name, &pfid);
1947 lee = (struct link_ea_entry *) ((char *)lee + reclen);
1948 if (mdt_fids_different_target(info, &pfid,
1949 mdt_object_fid(child)))
1956 int mdt_pack_encctx_in_reply(struct mdt_thread_info *info,
1957 struct mdt_object *child)
1959 struct lu_buf *buffer;
1960 struct mdt_body *repbody;
1961 struct req_capsule *pill = info->mti_pill;
1962 struct obd_export *exp = mdt_info_req(info)->rq_export;
1965 if (!exp_connect_encrypt(exp))
1968 if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
1969 req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0) {
1970 struct lu_attr la = { 0 };
1971 struct dt_object *dt = mdt_obj2dt(child);
1973 if (dt && dt->do_ops && dt->do_ops->do_attr_get)
1974 dt_attr_get(info->mti_env, mdt_obj2dt(child), &la);
1976 if (la.la_valid & LA_FLAGS && la.la_flags & LUSTRE_ENCRYPT_FL) {
1977 buffer = &info->mti_buf;
1979 /* fill reply buffer with encryption context now */
1981 req_capsule_get_size(pill, &RMF_FILE_ENCCTX,
1984 req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
1985 rc = mo_xattr_get(info->mti_env,
1986 mdt_object_child(child),
1988 LL_XATTR_NAME_ENCRYPTION_CONTEXT);
1991 "found encryption ctx of size %d for "DFID"\n",
1992 rc, PFID(mdt_object_fid(child)));
1994 repbody = req_capsule_server_get(pill,
1996 repbody->mbo_valid |= OBD_MD_ENCCTX;
1997 if (rc < buffer->lb_len)
1998 req_capsule_shrink(pill,
1999 &RMF_FILE_ENCCTX, rc,
2004 "encryption ctx not found for "DFID": rc = %d\n",
2005 PFID(mdt_object_fid(child)), rc);
2006 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,
2008 /* handling -ENOENT is important because it may
2009 * change object state in DNE env dropping
2010 * LOHA_EXISTS flag, it is important to return
2011 * that to the caller.
2012 * Check LU-13115 for details.
2018 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,