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) {
320 ucred->uc_fsuid = nodemap->nm_squash_uid;
321 ucred->uc_fsgid = nodemap->nm_squash_gid;
324 ucred->uc_fsuid = pud->pud_fsuid;
325 ucred->uc_fsgid = pud->pud_fsgid;
326 ucred->uc_cap = pud->pud_cap;
329 /* process root_squash here. */
330 mdt_root_squash(info, peernid);
332 ucred->uc_valid = UCRED_NEW;
333 ucred_set_jobid(info, ucred);
334 ucred_set_nid(info, ucred);
335 ucred_set_audit_enabled(info, ucred);
341 if (ucred->uc_ginfo) {
342 put_group_info(ucred->uc_ginfo);
343 ucred->uc_ginfo = NULL;
345 if (ucred->uc_identity) {
346 mdt_identity_put(mdt->mdt_identity_cache,
348 ucred->uc_identity = NULL;
356 * Check whether allow the client to set supplementary group IDs or not.
358 * \param[in] info pointer to the thread context
359 * \param[in] uc pointer to the RPC user descriptor
361 * \retval true if allow to set supplementary group IDs
362 * \retval false for other cases
364 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
368 /* 1. If identity_upcall is disabled,
369 * permit local client to do anything. */
370 if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
373 /* 2. If fail to get related identities, then forbid any client to
374 * set supplementary group IDs. */
375 if (uc->uc_identity == NULL)
378 /* 3. Check the permission in the identities. */
379 perm = mdt_identity_get_perm(uc->uc_identity,
380 mdt_info_req(info)->rq_peer.nid);
381 if (perm & CFS_SETGRP_PERM)
387 int mdt_check_ucred(struct mdt_thread_info *info)
389 struct ptlrpc_request *req = mdt_info_req(info);
390 struct mdt_device *mdt = info->mti_mdt;
391 struct ptlrpc_user_desc *pud = req->rq_user_desc;
392 struct lu_ucred *ucred = mdt_ucred(info);
393 struct md_identity *identity = NULL;
394 lnet_nid_t peernid = req->rq_peer.nid;
402 LASSERT(ucred != NULL);
403 if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
406 if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
409 /* sanity check: if we use strong authentication, we expect the
410 * uid which client claimed is true */
411 if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
412 req->rq_auth_uid != pud->pud_uid) {
413 CDEBUG(D_SEC, "local client %s: auth uid %u "
414 "while client claims %u:%u/%u:%u\n",
415 libcfs_nid2str(peernid), req->rq_auth_uid,
416 pud->pud_uid, pud->pud_gid,
417 pud->pud_fsuid, pud->pud_fsgid);
421 if (is_identity_get_disabled(mdt->mdt_identity_cache))
424 identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
425 if (IS_ERR(identity)) {
426 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
429 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
435 perm = mdt_identity_get_perm(identity, peernid);
436 /* find out the setuid/setgid attempt */
437 setuid = (pud->pud_uid != pud->pud_fsuid);
438 setgid = (pud->pud_gid != pud->pud_fsgid ||
439 pud->pud_gid != identity->mi_gid);
441 /* check permission of setuid */
442 if (setuid && !(perm & CFS_SETUID_PERM)) {
443 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
444 pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
445 GOTO(out, rc = -EACCES);
448 /* check permission of setgid */
449 if (setgid && !(perm & CFS_SETGID_PERM)) {
450 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
451 "from %s\n", pud->pud_uid, pud->pud_gid,
452 pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
453 libcfs_nid2str(peernid));
454 GOTO(out, rc = -EACCES);
460 mdt_identity_put(mdt->mdt_identity_cache, identity);
464 static int old_init_ucred_common(struct mdt_thread_info *info,
465 struct lu_nodemap *nodemap)
467 struct lu_ucred *uc = mdt_ucred(info);
468 struct mdt_device *mdt = info->mti_mdt;
469 struct md_identity *identity = NULL;
471 if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
472 /* deny access before we get identity ref */
473 if (nodemap->nmf_deny_unknown)
476 uc->uc_fsuid = nodemap->nm_squash_uid;
477 uc->uc_fsgid = nodemap->nm_squash_gid;
479 uc->uc_suppgids[0] = -1;
480 uc->uc_suppgids[1] = -1;
483 if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
484 identity = mdt_identity_get(mdt->mdt_identity_cache,
486 if (IS_ERR(identity)) {
487 if (unlikely(PTR_ERR(identity) == -EREMCHG ||
488 uc->uc_cap & CFS_CAP_FS_MASK)) {
491 CDEBUG(D_SEC, "Deny access without identity: "
492 "uid %u\n", uc->uc_fsuid);
497 uc->uc_identity = identity;
499 /* process root_squash here. */
500 mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
502 uc->uc_valid = UCRED_OLD;
503 ucred_set_jobid(info, uc);
504 ucred_set_nid(info, uc);
505 ucred_set_audit_enabled(info, uc);
512 static int old_init_ucred(struct mdt_thread_info *info,
513 struct mdt_body *body)
515 struct lu_ucred *uc = mdt_ucred(info);
516 struct lu_nodemap *nodemap;
520 nodemap = nodemap_get_from_exp(info->mti_exp);
522 RETURN(PTR_ERR(nodemap));
524 body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
525 NODEMAP_CLIENT_TO_FS, body->mbo_uid);
526 body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
527 NODEMAP_CLIENT_TO_FS, body->mbo_gid);
528 body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
529 NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
530 body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
531 NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
534 uc->uc_valid = UCRED_INVALID;
535 uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
536 uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
537 uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
538 uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
539 uc->uc_suppgids[0] = body->mbo_suppgid;
540 uc->uc_suppgids[1] = -1;
542 uc->uc_cap = body->mbo_capability;
544 rc = old_init_ucred_common(info, nodemap);
545 nodemap_putref(nodemap);
550 static int old_init_ucred_reint(struct mdt_thread_info *info)
552 struct lu_ucred *uc = mdt_ucred(info);
553 struct lu_nodemap *nodemap;
557 nodemap = nodemap_get_from_exp(info->mti_exp);
559 RETURN(PTR_ERR(nodemap));
563 uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
564 NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
565 uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
566 NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
568 uc->uc_valid = UCRED_INVALID;
569 uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
570 uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
573 rc = old_init_ucred_common(info, nodemap);
574 nodemap_putref(nodemap);
579 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
580 struct mdt_body *body)
582 struct ptlrpc_request *req = mdt_info_req(info);
583 struct lu_ucred *uc = mdt_ucred(info);
586 if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
589 mdt_exit_ucred(info);
591 if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
592 return old_init_ucred(info, body);
594 return new_init_ucred(info, BODY_INIT, body);
597 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
599 return __mdt_init_ucred(info, body);
602 int mdt_init_ucred_reint(struct mdt_thread_info *info)
604 struct ptlrpc_request *req = mdt_info_req(info);
605 struct lu_ucred *uc = mdt_ucred(info);
606 struct md_attr *ma = &info->mti_attr;
609 if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
612 /* LU-5564: for normal close request, skip permission check */
613 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
614 !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
615 uc->uc_cap |= CFS_CAP_FS_MASK;
617 mdt_exit_ucred(info);
619 if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
620 return old_init_ucred_reint(info);
622 return new_init_ucred(info, REC_INIT, NULL);
625 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
626 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
628 const struct lov_ost_data_v1 *lod;
629 __u32 lmm_magic = le32_to_cpu(lmm->lmm_magic);
633 if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
636 CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
637 POSTID(&lmm->lmm_oi), lmm_magic,
638 le32_to_cpu(lmm->lmm_pattern));
640 /* No support for compount layouts yet */
641 if (lmm_magic != LOV_MAGIC_V1 && lmm_magic != LOV_MAGIC_V3)
644 count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
645 CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
646 le32_to_cpu(lmm->lmm_stripe_size), count);
648 /* If it's a directory or a released file, then there are
649 * no actual objects to print, so bail out. */
650 if (valid & OBD_MD_FLDIREA ||
651 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
654 LASSERT(count <= LOV_MAX_STRIPE_COUNT);
655 for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
658 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
659 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
660 i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
664 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
666 const struct lmv_mds_md_v1 *lmm1;
667 const struct lmv_foreign_md *lfm;
670 if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
673 /* foreign LMV case */
674 lfm = &lmv->lmv_foreign_md;
675 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
677 "foreign magic 0x%08X, length %u, type %u, flags %u, value '%.*s'\n",
678 le32_to_cpu(lfm->lfm_magic),
679 le32_to_cpu(lfm->lfm_length),
680 le32_to_cpu(lfm->lfm_type),
681 le32_to_cpu(lfm->lfm_flags),
682 le32_to_cpu(lfm->lfm_length), lfm->lfm_value);
686 lmm1 = &lmv->lmv_md_v1;
688 "magic 0x%08X, master %#X stripe_count %#x hash_type %#x\n",
689 le32_to_cpu(lmm1->lmv_magic),
690 le32_to_cpu(lmm1->lmv_master_mdt_index),
691 le32_to_cpu(lmm1->lmv_stripe_count),
692 le32_to_cpu(lmm1->lmv_hash_type));
694 if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
697 for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
700 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
701 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
705 /* Shrink and/or grow reply buffers */
706 int mdt_fix_reply(struct mdt_thread_info *info)
708 struct req_capsule *pill = info->mti_pill;
709 struct mdt_body *body;
710 int md_size, md_packed = 0;
715 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
716 LASSERT(body != NULL);
718 if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
720 md_size = body->mbo_eadatasize;
724 acl_size = body->mbo_aclsize;
726 /* this replay - not send info to client */
727 if (info->mti_spec.no_create) {
732 CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
737 &RMF_ACL, or &RMF_LOGCOOKIES
738 (optional) &RMF_CAPA1,
739 (optional) &RMF_CAPA2,
740 (optional) something else
743 /* MDT_MD buffer may be bigger than packed value, let's shrink all
744 * buffers before growing it */
745 if (info->mti_big_lmm_used) {
746 /* big_lmm buffer may be used even without packing the result
747 * into reply, just for internal server needs */
748 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
749 md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
752 /* free big lmm if md_size is not needed */
753 if (md_size == 0 || md_packed == 0) {
754 info->mti_big_lmm_used = 0;
756 /* buffer must be allocated separately */
757 LASSERT(info->mti_attr.ma_lmm !=
758 req_capsule_server_get(pill, &RMF_MDT_MD));
759 req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
761 } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
762 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
765 if (info->mti_big_acl_used) {
767 info->mti_big_acl_used = 0;
769 req_capsule_shrink(pill, &RMF_ACL, 0, RCL_SERVER);
770 } else if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER)) {
771 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
772 } else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) {
773 req_capsule_shrink(pill, &RMF_LOGCOOKIES, acl_size, RCL_SERVER);
776 /* Shrink optional SECCTX buffer if it is not used */
777 if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
778 req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0 &&
779 !(body->mbo_valid & OBD_MD_SECCTX))
780 req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0, RCL_SERVER);
782 /* Shrink optional ENCCTX buffer if it is not used */
783 if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
784 req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0 &&
785 !(body->mbo_valid & OBD_MD_ENCCTX))
786 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0, RCL_SERVER);
789 * Some more field should be shrinked if needed.
790 * This should be done by those who added fields to reply message.
793 /* Grow MD buffer if needed finally */
794 if (info->mti_big_lmm_used) {
797 LASSERT(md_size > md_packed);
798 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
799 md_size - md_packed);
800 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
802 /* we can't answer with proper LOV EA, drop flags,
803 * the rc is also returned so this request is
804 * considered as failed */
805 body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
806 /* don't return transno along with error */
807 lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
809 /* now we need to pack right LOV/LMV EA */
810 lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
811 if (info->mti_attr.ma_valid & MA_LOV) {
812 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
814 info->mti_attr.ma_lmm_size);
815 memcpy(lmm, info->mti_attr.ma_lmm,
816 info->mti_attr.ma_lmm_size);
817 } else if (info->mti_attr.ma_valid & MA_LMV) {
818 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
820 info->mti_attr.ma_lmv_size);
821 memcpy(lmm, info->mti_attr.ma_lmv,
822 info->mti_attr.ma_lmv_size);
826 /* update mdt_max_mdsize so clients will be aware about that */
827 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
828 info->mti_mdt->mdt_max_mdsize =
829 info->mti_attr.ma_lmm_size;
830 info->mti_big_lmm_used = 0;
833 if (info->mti_big_acl_used) {
834 CDEBUG(D_INFO, "Enlarge reply ACL buffer to %d bytes\n",
837 rc = req_capsule_server_grow(pill, &RMF_ACL, acl_size);
839 body->mbo_valid &= ~OBD_MD_FLACL;
841 void *acl = req_capsule_server_get(pill, &RMF_ACL);
843 memcpy(acl, info->mti_big_acl, acl_size);
846 info->mti_big_acl_used = 0;
853 /* if object is dying, pack the lov/llog data,
854 * parameter info->mti_attr should be valid at this point!
855 * Also implements RAoLU policy */
856 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
859 struct mdt_body *repbody = NULL;
860 const struct lu_attr *la = &ma->ma_attr;
861 struct coordinator *cdt = &info->mti_mdt->mdt_coordinator;
864 struct hsm_action_item hai = {
865 .hai_len = sizeof(hai),
866 .hai_action = HSMA_REMOVE,
867 .hai_extent.length = -1,
875 if (mdt_info_req(info) != NULL) {
876 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
877 LASSERT(repbody != NULL);
879 CDEBUG(D_INFO, "not running in a request/reply context\n");
882 if ((ma->ma_valid & MA_INODE) && repbody != NULL)
883 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
885 if (ma->ma_valid & MA_LOV) {
886 CERROR("No need in LOV EA upon unlink\n");
890 repbody->mbo_eadatasize = 0;
892 /* Only check unlinked and archived if RAoLU and upon last close */
893 if (!cdt->cdt_remove_archive_on_last_unlink ||
894 atomic_read(&mo->mot_open_count) != 0)
897 /* mdt_attr_get_complex will clear ma_valid, so check here first */
898 if ((ma->ma_valid & MA_INODE) && (ma->ma_attr.la_nlink != 0))
901 if ((ma->ma_valid & MA_HSM) && (!(ma->ma_hsm.mh_flags & HS_EXISTS)))
904 need |= (MA_INODE | MA_HSM) & ~ma->ma_valid;
906 /* ma->ma_valid is missing either MA_INODE, MA_HSM, or both,
907 * try setting them */
909 rc = mdt_attr_get_complex(info, mo, ma);
911 CERROR("%s: unable to fetch missing attributes of"
912 DFID": rc=%d\n", mdt_obd_name(info->mti_mdt),
913 PFID(mdt_object_fid(mo)), rc);
917 if (need & MA_INODE) {
918 if (ma->ma_valid & MA_INODE) {
919 if (ma->ma_attr.la_nlink != 0)
927 if (ma->ma_valid & MA_HSM) {
928 if (!(ma->ma_hsm.mh_flags & HS_EXISTS))
936 /* RAoLU policy is active, last close on file has occured,
937 * file is unlinked, file is archived, so create remove request
939 * If CDT is not running, requests will be logged for later. */
940 if (ma->ma_hsm.mh_arch_id != 0)
941 archive_id = ma->ma_hsm.mh_arch_id;
943 archive_id = cdt->cdt_default_archive_id;
945 hai.hai_fid = *mdt_object_fid(mo);
947 rc = mdt_agent_record_add(info->mti_env, info->mti_mdt, archive_id, 0,
950 CERROR("%s: unable to add HSM remove request for "DFID
951 ": rc=%d\n", mdt_obd_name(info->mti_mdt),
952 PFID(mdt_object_fid(mo)), rc);
957 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
963 if (in & MDS_ATTR_MODE)
965 if (in & MDS_ATTR_UID)
967 if (in & MDS_ATTR_GID)
969 if (in & MDS_ATTR_SIZE)
971 if (in & MDS_ATTR_BLOCKS)
973 if (in & MDS_ATTR_ATIME_SET)
975 if (in & MDS_ATTR_CTIME_SET)
977 if (in & MDS_ATTR_MTIME_SET)
979 if (in & MDS_ATTR_ATTR_FLAG)
981 if (in & MDS_ATTR_KILL_SUID)
983 if (in & MDS_ATTR_KILL_SGID)
985 if (in & MDS_ATTR_PROJID)
987 if (in & MDS_ATTR_LSIZE)
989 if (in & MDS_ATTR_LBLOCKS)
992 if (in & MDS_ATTR_FROM_OPEN)
993 rr->rr_flags |= MRF_OPEN_TRUNC;
994 if (in & MDS_ATTR_OVERRIDE)
995 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
996 if (in & MDS_ATTR_FORCE)
997 ma->ma_attr_flags |= MDS_PERM_BYPASS;
999 in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | MDS_ATTR_PROJID |
1000 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
1001 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
1002 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
1003 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
1004 MDS_ATTR_FROM_OPEN | MDS_ATTR_LSIZE | MDS_ATTR_LBLOCKS |
1007 CDEBUG(D_INFO, "Unknown attr bits: %#llx\n", in);
1013 int mdt_name_unpack(struct req_capsule *pill,
1014 const struct req_msg_field *field,
1016 enum mdt_name_flags flags)
1018 ln->ln_name = req_capsule_client_get(pill, field);
1019 ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
1021 if (!lu_name_is_valid(ln)) {
1028 if ((flags & MNF_FIX_ANON) &&
1029 ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
1030 /* Newer (3.x) kernels use a name of "/" for the
1031 * "anonymous" disconnected dentries from NFS
1032 * filehandle conversion. See d_obtain_alias(). */
1040 static int mdt_file_secctx_unpack(struct req_capsule *pill,
1041 const char **secctx_name,
1042 void **secctx, size_t *secctx_size)
1047 *secctx_name = NULL;
1051 if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
1052 !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
1055 name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
1060 if (name_size > XATTR_NAME_MAX + 1)
1063 name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1064 if (strnlen(name, name_size) != name_size - 1)
1067 if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
1068 !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
1071 *secctx_name = name;
1072 *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
1073 *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
1078 static int mdt_file_encctx_unpack(struct req_capsule *pill,
1079 void **encctx, size_t *encctx_size)
1084 if (!exp_connect_encrypt(pill->rc_req->rq_export))
1087 if (!req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_CLIENT) ||
1088 !req_capsule_field_present(pill, &RMF_FILE_ENCCTX, RCL_CLIENT))
1091 *encctx_size = req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_CLIENT);
1092 if (*encctx_size == 0)
1095 *encctx = req_capsule_client_get(pill, &RMF_FILE_ENCCTX);
1100 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
1102 struct lu_ucred *uc = mdt_ucred(info);
1103 struct md_attr *ma = &info->mti_attr;
1104 struct lu_attr *la = &ma->ma_attr;
1105 struct req_capsule *pill = info->mti_pill;
1106 struct mdt_reint_record *rr = &info->mti_rr;
1107 struct mdt_rec_setattr *rec;
1108 struct lu_nodemap *nodemap;
1112 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1113 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1117 /* This prior initialization is needed for old_init_ucred_reint() */
1118 uc->uc_fsuid = rec->sa_fsuid;
1119 uc->uc_fsgid = rec->sa_fsgid;
1120 uc->uc_cap = rec->sa_cap;
1121 uc->uc_suppgids[0] = rec->sa_suppgid;
1122 uc->uc_suppgids[1] = -1;
1124 rr->rr_fid1 = &rec->sa_fid;
1125 la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
1126 la->la_mode = rec->sa_mode;
1127 la->la_flags = rec->sa_attr_flags;
1129 nodemap = nodemap_get_from_exp(info->mti_exp);
1130 if (IS_ERR(nodemap))
1131 RETURN(PTR_ERR(nodemap));
1133 la->la_uid = nodemap_map_id(nodemap, NODEMAP_UID,
1134 NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1135 la->la_gid = nodemap_map_id(nodemap, NODEMAP_GID,
1136 NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1137 la->la_projid = rec->sa_projid;
1138 nodemap_putref(nodemap);
1140 la->la_size = rec->sa_size;
1141 la->la_blocks = rec->sa_blocks;
1142 la->la_ctime = rec->sa_ctime;
1143 la->la_atime = rec->sa_atime;
1144 la->la_mtime = rec->sa_mtime;
1145 ma->ma_valid = MA_INODE;
1147 ma->ma_attr_flags |= rec->sa_bias & (MDS_CLOSE_INTENT |
1148 MDS_DATA_MODIFIED | MDS_TRUNC_KEEP_LEASE |
1153 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
1155 struct req_capsule *pill = info->mti_pill;
1156 struct mdt_ioepoch *ioepoch;
1159 if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1160 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1164 if (ioepoch == NULL)
1167 info->mti_open_handle = ioepoch->mio_open_handle;
1172 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1173 struct req_capsule *pill = info->mti_pill;
1175 if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1176 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1177 if (info->mti_dlm_req == NULL)
1184 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1186 struct mdt_reint_record *rr = &info->mti_rr;
1187 struct md_attr *ma = &info->mti_attr;
1188 struct req_capsule *pill = info->mti_pill;
1192 rc = mdt_setattr_unpack_rec(info);
1196 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1197 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1198 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1201 if (rr->rr_eadatalen > 0) {
1202 const struct lmv_user_md *lum;
1204 lum = rr->rr_eadata;
1205 /* Sigh ma_valid(from req) does not indicate whether
1206 * it will set LOV/LMV EA, so we have to check magic */
1207 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1208 ma->ma_valid |= MA_LMV;
1209 ma->ma_lmv = (void *)rr->rr_eadata;
1210 ma->ma_lmv_size = rr->rr_eadatalen;
1212 ma->ma_valid |= MA_LOV;
1213 ma->ma_lmm = (void *)rr->rr_eadata;
1214 ma->ma_lmm_size = rr->rr_eadatalen;
1219 rc = mdt_dlmreq_unpack(info);
1223 static int mdt_close_intent_unpack(struct mdt_thread_info *info)
1225 struct md_attr *ma = &info->mti_attr;
1226 struct req_capsule *pill = info->mti_pill;
1229 if (!(ma->ma_attr_flags & MDS_CLOSE_INTENT))
1232 req_capsule_extend(pill, &RQF_MDS_CLOSE_INTENT);
1234 if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1235 req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1241 int mdt_close_unpack(struct mdt_thread_info *info)
1246 rc = mdt_close_handle_unpack(info);
1250 rc = mdt_setattr_unpack_rec(info);
1254 rc = mdt_close_intent_unpack(info);
1258 RETURN(mdt_init_ucred_reint(info));
1261 static int mdt_create_unpack(struct mdt_thread_info *info)
1263 struct lu_ucred *uc = mdt_ucred(info);
1264 struct mdt_rec_create *rec;
1265 struct lu_attr *attr = &info->mti_attr.ma_attr;
1266 struct mdt_reint_record *rr = &info->mti_rr;
1267 struct req_capsule *pill = info->mti_pill;
1268 struct md_op_spec *sp = &info->mti_spec;
1273 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1274 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1278 /* This prior initialization is needed for old_init_ucred_reint() */
1279 uc->uc_fsuid = rec->cr_fsuid;
1280 uc->uc_fsgid = rec->cr_fsgid;
1281 uc->uc_cap = rec->cr_cap;
1282 uc->uc_suppgids[0] = rec->cr_suppgid1;
1283 uc->uc_suppgids[1] = -1;
1284 uc->uc_umask = rec->cr_umask;
1286 rr->rr_fid1 = &rec->cr_fid1;
1287 rr->rr_fid2 = &rec->cr_fid2;
1288 attr->la_mode = rec->cr_mode;
1289 attr->la_rdev = rec->cr_rdev;
1290 attr->la_uid = rec->cr_fsuid;
1291 attr->la_gid = rec->cr_fsgid;
1292 attr->la_ctime = rec->cr_time;
1293 attr->la_mtime = rec->cr_time;
1294 attr->la_atime = rec->cr_time;
1295 attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1296 LA_CTIME | LA_MTIME | LA_ATIME;
1297 memset(&sp->u, 0, sizeof(sp->u));
1298 sp->sp_cr_flags = get_mrc_cr_flags(rec);
1300 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1304 if (S_ISLNK(attr->la_mode)) {
1305 const char *tgt = NULL;
1307 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1308 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1309 tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1310 sp->u.sp_symname = tgt;
1315 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1316 if (S_ISDIR(attr->la_mode) &&
1317 req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1318 sp->u.sp_ea.eadata =
1319 req_capsule_client_get(pill, &RMF_EADATA);
1320 sp->u.sp_ea.eadatalen =
1321 req_capsule_get_size(pill, &RMF_EADATA,
1323 sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1327 rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1328 &sp->sp_cr_file_secctx,
1329 &sp->sp_cr_file_secctx_size);
1333 rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
1334 &sp->sp_cr_file_encctx_size);
1338 rc = req_check_sepol(pill);
1342 rc = mdt_dlmreq_unpack(info);
1346 static int mdt_link_unpack(struct mdt_thread_info *info)
1348 struct lu_ucred *uc = mdt_ucred(info);
1349 struct mdt_rec_link *rec;
1350 struct lu_attr *attr = &info->mti_attr.ma_attr;
1351 struct mdt_reint_record *rr = &info->mti_rr;
1352 struct req_capsule *pill = info->mti_pill;
1357 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1358 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1362 /* This prior initialization is needed for old_init_ucred_reint() */
1363 uc->uc_fsuid = rec->lk_fsuid;
1364 uc->uc_fsgid = rec->lk_fsgid;
1365 uc->uc_cap = rec->lk_cap;
1366 uc->uc_suppgids[0] = rec->lk_suppgid1;
1367 uc->uc_suppgids[1] = rec->lk_suppgid2;
1369 attr->la_uid = rec->lk_fsuid;
1370 attr->la_gid = rec->lk_fsgid;
1371 rr->rr_fid1 = &rec->lk_fid1;
1372 rr->rr_fid2 = &rec->lk_fid2;
1373 attr->la_ctime = rec->lk_time;
1374 attr->la_mtime = rec->lk_time;
1375 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1377 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1381 rc = req_check_sepol(pill);
1385 rc = mdt_dlmreq_unpack(info);
1390 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1392 struct lu_ucred *uc = mdt_ucred(info);
1393 struct mdt_rec_unlink *rec;
1394 struct lu_attr *attr = &info->mti_attr.ma_attr;
1395 struct mdt_reint_record *rr = &info->mti_rr;
1396 struct req_capsule *pill = info->mti_pill;
1401 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1402 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1406 /* This prior initialization is needed for old_init_ucred_reint() */
1407 uc->uc_fsuid = rec->ul_fsuid;
1408 uc->uc_fsgid = rec->ul_fsgid;
1409 uc->uc_cap = rec->ul_cap;
1410 uc->uc_suppgids[0] = rec->ul_suppgid1;
1411 uc->uc_suppgids[1] = -1;
1413 attr->la_uid = rec->ul_fsuid;
1414 attr->la_gid = rec->ul_fsgid;
1415 rr->rr_fid1 = &rec->ul_fid1;
1416 rr->rr_fid2 = &rec->ul_fid2;
1417 attr->la_ctime = rec->ul_time;
1418 attr->la_mtime = rec->ul_time;
1419 attr->la_mode = rec->ul_mode;
1420 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1422 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1426 info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1428 rc = req_check_sepol(pill);
1432 rc = mdt_dlmreq_unpack(info);
1436 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1438 info->mti_spec.sp_rm_entry = 1;
1439 return mdt_unlink_unpack(info);
1442 static int mdt_rename_unpack(struct mdt_thread_info *info)
1444 struct lu_ucred *uc = mdt_ucred(info);
1445 struct mdt_rec_rename *rec;
1446 struct lu_attr *attr = &info->mti_attr.ma_attr;
1447 struct mdt_reint_record *rr = &info->mti_rr;
1448 struct req_capsule *pill = info->mti_pill;
1449 struct md_op_spec *spec = &info->mti_spec;
1454 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1455 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1459 /* This prior initialization is needed for old_init_ucred_reint() */
1460 uc->uc_fsuid = rec->rn_fsuid;
1461 uc->uc_fsgid = rec->rn_fsgid;
1462 uc->uc_cap = rec->rn_cap;
1463 uc->uc_suppgids[0] = rec->rn_suppgid1;
1464 uc->uc_suppgids[1] = rec->rn_suppgid2;
1466 attr->la_uid = rec->rn_fsuid;
1467 attr->la_gid = rec->rn_fsgid;
1468 rr->rr_fid1 = &rec->rn_fid1;
1469 rr->rr_fid2 = &rec->rn_fid2;
1470 attr->la_ctime = rec->rn_time;
1471 attr->la_mtime = rec->rn_time;
1472 /* rename_tgt contains the mode already */
1473 attr->la_mode = rec->rn_mode;
1474 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1476 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1480 rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1484 spec->no_create = !!req_is_replay(mdt_info_req(info));
1486 rc = req_check_sepol(pill);
1490 rc = mdt_dlmreq_unpack(info);
1495 static int mdt_migrate_unpack(struct mdt_thread_info *info)
1497 struct lu_ucred *uc = mdt_ucred(info);
1498 struct mdt_rec_rename *rec;
1499 struct lu_attr *attr = &info->mti_attr.ma_attr;
1500 struct mdt_reint_record *rr = &info->mti_rr;
1501 struct req_capsule *pill = info->mti_pill;
1502 struct md_op_spec *spec = &info->mti_spec;
1507 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1508 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1512 /* This prior initialization is needed for old_init_ucred_reint() */
1513 uc->uc_fsuid = rec->rn_fsuid;
1514 uc->uc_fsgid = rec->rn_fsgid;
1515 uc->uc_cap = rec->rn_cap;
1516 uc->uc_suppgids[0] = rec->rn_suppgid1;
1517 uc->uc_suppgids[1] = rec->rn_suppgid2;
1519 attr->la_uid = rec->rn_fsuid;
1520 attr->la_gid = rec->rn_fsgid;
1521 rr->rr_fid1 = &rec->rn_fid1;
1522 rr->rr_fid2 = &rec->rn_fid2;
1523 attr->la_ctime = rec->rn_time;
1524 attr->la_mtime = rec->rn_time;
1525 /* rename_tgt contains the mode already */
1526 attr->la_mode = rec->rn_mode;
1527 attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1528 spec->sp_cr_flags = 0;
1530 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1534 if (rec->rn_bias & MDS_CLOSE_MIGRATE) {
1535 rc = mdt_close_handle_unpack(info);
1539 spec->sp_migrate_close = 1;
1541 spec->sp_migrate_close = 0;
1543 spec->sp_migrate_nsonly = 0;
1545 /* lustre version > 2.11 migration packs lum */
1546 if (req_capsule_has_field(pill, &RMF_EADATA, RCL_CLIENT)) {
1547 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1548 rr->rr_eadatalen = req_capsule_get_size(pill,
1552 if (rr->rr_eadatalen > 0) {
1553 rr->rr_eadata = req_capsule_client_get(pill,
1555 spec->u.sp_ea.eadatalen = rr->rr_eadatalen;
1556 spec->u.sp_ea.eadata = rr->rr_eadata;
1557 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1560 /* old client doesn't provide lum. */
1561 RETURN(-EOPNOTSUPP);
1565 spec->no_create = !!req_is_replay(mdt_info_req(info));
1567 rc = mdt_dlmreq_unpack(info);
1573 * please see comment above LOV_MAGIC_V1_DEFINED
1575 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata)
1577 struct lov_user_md_v1 *v1 = eadata;
1581 if (unlikely(req_is_replay(mdt_info_req(info)))) {
1582 if ((v1->lmm_magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
1583 v1->lmm_magic |= LOV_MAGIC_DEFINED;
1584 else if ((v1->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
1585 __swab32(LOV_MAGIC_MAGIC))
1586 v1->lmm_magic |= __swab32(LOV_MAGIC_DEFINED);
1590 static int mdt_open_unpack(struct mdt_thread_info *info)
1592 struct lu_ucred *uc = mdt_ucred(info);
1593 struct mdt_rec_create *rec;
1594 struct lu_attr *attr = &info->mti_attr.ma_attr;
1595 struct req_capsule *pill = info->mti_pill;
1596 struct mdt_reint_record *rr = &info->mti_rr;
1597 struct ptlrpc_request *req = mdt_info_req(info);
1598 struct md_op_spec *sp = &info->mti_spec;
1602 BUILD_BUG_ON(sizeof(struct mdt_rec_create) !=
1603 sizeof(struct mdt_rec_reint));
1604 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1608 /* This prior initialization is needed for old_init_ucred_reint() */
1609 uc->uc_fsuid = rec->cr_fsuid;
1610 uc->uc_fsgid = rec->cr_fsgid;
1611 uc->uc_cap = rec->cr_cap;
1612 uc->uc_suppgids[0] = rec->cr_suppgid1;
1613 uc->uc_suppgids[1] = rec->cr_suppgid2;
1614 uc->uc_umask = rec->cr_umask;
1616 rr->rr_fid1 = &rec->cr_fid1;
1617 rr->rr_fid2 = &rec->cr_fid2;
1618 rr->rr_open_handle = &rec->cr_open_handle_old;
1619 attr->la_mode = rec->cr_mode;
1620 attr->la_rdev = rec->cr_rdev;
1621 attr->la_uid = rec->cr_fsuid;
1622 attr->la_gid = rec->cr_fsgid;
1623 attr->la_ctime = rec->cr_time;
1624 attr->la_mtime = rec->cr_time;
1625 attr->la_atime = rec->cr_time;
1626 attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID |
1627 LA_CTIME | LA_MTIME | LA_ATIME;
1628 memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1629 info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1630 /* Do not trigger ASSERTION if client miss to set such flags. */
1631 if (unlikely(info->mti_spec.sp_cr_flags == 0))
1634 info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1636 mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1638 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1639 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1642 if (rr->rr_eadatalen > 0) {
1643 rr->rr_eadata = req_capsule_client_get(pill,
1645 sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1646 sp->u.sp_ea.eadata = rr->rr_eadata;
1647 sp->sp_archive_id = rec->cr_archive_id;
1648 sp->no_create = !!req_is_replay(req);
1649 mdt_fix_lov_magic(info, rr->rr_eadata);
1653 * Client default md_size may be 0 right after client start,
1654 * until all osc are connected, set here just some reasonable
1655 * value to prevent misbehavior.
1657 if (rr->rr_eadatalen == 0 &&
1658 !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1659 rr->rr_eadatalen = MIN_MD_SIZE;
1662 rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1663 &sp->sp_cr_file_secctx,
1664 &sp->sp_cr_file_secctx_size);
1668 rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
1669 &sp->sp_cr_file_encctx_size);
1673 rc = req_check_sepol(pill);
1680 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1682 struct mdt_reint_record *rr = &info->mti_rr;
1683 struct lu_ucred *uc = mdt_ucred(info);
1684 struct lu_attr *attr = &info->mti_attr.ma_attr;
1685 struct req_capsule *pill = info->mti_pill;
1686 struct mdt_rec_setxattr *rec;
1691 BUILD_BUG_ON(sizeof(struct mdt_rec_setxattr) !=
1692 sizeof(struct mdt_rec_reint));
1693 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1697 /* This prior initialization is needed for old_init_ucred_reint() */
1698 uc->uc_fsuid = rec->sx_fsuid;
1699 uc->uc_fsgid = rec->sx_fsgid;
1700 uc->uc_cap = rec->sx_cap;
1701 uc->uc_suppgids[0] = rec->sx_suppgid1;
1702 uc->uc_suppgids[1] = -1;
1704 rr->rr_opcode = rec->sx_opcode;
1705 rr->rr_fid1 = &rec->sx_fid;
1706 attr->la_valid = rec->sx_valid;
1707 attr->la_ctime = rec->sx_time;
1708 attr->la_size = rec->sx_size;
1709 attr->la_flags = rec->sx_flags;
1711 rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1715 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1716 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1719 if (rr->rr_eadatalen > info->mti_mdt->mdt_max_ea_size)
1722 if (rr->rr_eadatalen > 0) {
1723 rr->rr_eadata = req_capsule_client_get(pill,
1725 if (rr->rr_eadata == NULL)
1728 rr->rr_eadata = NULL;
1730 } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1731 CDEBUG(D_INFO, "no xattr data supplied\n");
1735 rc = req_check_sepol(pill);
1739 if (mdt_dlmreq_unpack(info) < 0)
1745 static int mdt_resync_unpack(struct mdt_thread_info *info)
1747 struct req_capsule *pill = info->mti_pill;
1748 struct mdt_reint_record *rr = &info->mti_rr;
1749 struct lu_ucred *uc = mdt_ucred(info);
1750 struct mdt_rec_resync *rec;
1753 BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1754 rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1758 /* This prior initialization is needed for old_init_ucred_reint() */
1759 uc->uc_fsuid = rec->rs_fsuid;
1760 uc->uc_fsgid = rec->rs_fsgid;
1761 uc->uc_cap = rec->rs_cap;
1763 rr->rr_fid1 = &rec->rs_fid;
1764 rr->rr_mirror_id = rec->rs_mirror_id;
1766 /* cookie doesn't need to be swapped but it has been swapped
1767 * in lustre_swab_mdt_rec_reint() as rr_mtime, so here it needs
1769 if (ptlrpc_req_need_swab(mdt_info_req(info)))
1770 __swab64s(&rec->rs_lease_handle.cookie);
1771 rr->rr_lease_handle = &rec->rs_lease_handle;
1773 RETURN(mdt_dlmreq_unpack(info));
1776 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1778 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1779 [REINT_SETATTR] = mdt_setattr_unpack,
1780 [REINT_CREATE] = mdt_create_unpack,
1781 [REINT_LINK] = mdt_link_unpack,
1782 [REINT_UNLINK] = mdt_unlink_unpack,
1783 [REINT_RENAME] = mdt_rename_unpack,
1784 [REINT_OPEN] = mdt_open_unpack,
1785 [REINT_SETXATTR] = mdt_setxattr_unpack,
1786 [REINT_RMENTRY] = mdt_rmentry_unpack,
1787 [REINT_MIGRATE] = mdt_migrate_unpack,
1788 [REINT_RESYNC] = mdt_resync_unpack,
1791 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1796 memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1797 if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1798 info->mti_rr.rr_opcode = op;
1799 rc = mdt_reint_unpackers[op](info);
1801 CERROR("Unexpected opcode %d\n", op);
1807 int mdt_pack_secctx_in_reply(struct mdt_thread_info *info,
1808 struct mdt_object *child)
1811 struct lu_buf *buffer;
1812 struct mdt_body *repbody;
1813 struct req_capsule *pill = info->mti_pill;
1816 if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
1817 req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0) {
1819 req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1820 buffer = &info->mti_buf;
1822 /* fill reply buffer with security context now */
1823 buffer->lb_len = req_capsule_get_size(pill, &RMF_FILE_SECCTX,
1825 buffer->lb_buf = req_capsule_server_get(info->mti_pill,
1827 rc = mo_xattr_get(info->mti_env, mdt_object_child(child),
1828 buffer, secctx_name);
1831 "found security context of size %d for "DFID"\n",
1832 rc, PFID(mdt_object_fid(child)));
1834 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1835 repbody->mbo_valid |= OBD_MD_SECCTX;
1836 if (rc < buffer->lb_len)
1837 req_capsule_shrink(pill, &RMF_FILE_SECCTX, rc,
1842 "security context not found for "DFID": rc = %d\n",
1843 PFID(mdt_object_fid(child)), rc);
1844 req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0,
1846 /* handling -ENOENT is important because it may change
1847 * object state in DNE env dropping LOHA_EXISTS flag,
1848 * it is important to return that to the caller.
1849 * Check LU-13115 for details.
1858 /* check whether two FIDs belong to different MDT. */
1859 static int mdt_fids_different_target(struct mdt_thread_info *info,
1860 const struct lu_fid *fid1,
1861 const struct lu_fid *fid2)
1863 const struct lu_env *env = info->mti_env;
1864 struct mdt_device *mdt = info->mti_mdt;
1865 struct lu_seq_range *range = &info->mti_range;
1866 struct seq_server_site *ss;
1867 __u32 index1, index2;
1870 if (fid_seq(fid1) == fid_seq(fid2))
1873 ss = mdt->mdt_lu_dev.ld_site->ld_seq_site;
1875 range->lsr_flags = LU_SEQ_RANGE_MDT;
1876 rc = fld_server_lookup(env, ss->ss_server_fld, fid1->f_seq, range);
1880 index1 = range->lsr_index;
1882 rc = fld_server_lookup(env, ss->ss_server_fld, fid2->f_seq, range);
1886 index2 = range->lsr_index;
1888 return index1 != index2;
1891 static bool mdt_object_is_shard(struct mdt_thread_info *info,
1892 struct mdt_object *obj)
1894 struct lmv_mds_md_v1 *lmv = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
1898 if (!mdt_object_exists(obj))
1901 if (!S_ISDIR(lu_object_attr(&obj->mot_obj)))
1905 buf.lb_len = sizeof(*lmv);
1906 rc = mo_xattr_get(info->mti_env, mdt_object_child(obj), &buf,
1911 return lmv->lmv_magic == cpu_to_le32(LMV_MAGIC_STRIPE);
1915 * Check whether \a child is remote object on \a parent.
1917 * \param[in] info thread environment
1918 * \param[in] parent parent object, it's the same as child object in
1920 * \param[in] child child object
1922 * \retval 1 is remote object.
1923 * \retval 0 isn't remote object.
1924 * \retval < 1 error code
1926 int mdt_is_remote_object(struct mdt_thread_info *info,
1927 struct mdt_object *parent,
1928 struct mdt_object *child)
1930 struct lu_buf *buf = &info->mti_big_buf;
1931 struct linkea_data ldata = { NULL };
1932 struct link_ea_header *leh;
1933 struct link_ea_entry *lee;
1934 struct lu_name name;
1942 if (fid_is_root(mdt_object_fid(child)))
1945 if (likely(parent != child)) {
1946 if (mdt_object_remote(parent) ^ mdt_object_remote(child)) {
1947 /* don't treat shard as remote object, otherwise client
1948 * need to revalidate shards all the time.
1950 if (mdt_object_is_shard(info, child))
1955 if (!mdt_object_remote(parent) && !mdt_object_remote(child))
1958 rc = mdt_fids_different_target(info, mdt_object_fid(parent),
1959 mdt_object_fid(child));
1963 /* client < 2.13.52 getattr_by_fid parent and child are the same */
1964 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
1969 rc = mdt_links_read(info, child, &ldata);
1970 /* can't read linkea, just assume it's remote object */
1971 if (rc == -ENOENT || rc == -ENODATA)
1977 lee = (struct link_ea_entry *)(leh + 1);
1978 for (i = 0; i < leh->leh_reccount; i++) {
1979 linkea_entry_unpack(lee, &reclen, &name, &pfid);
1980 lee = (struct link_ea_entry *) ((char *)lee + reclen);
1981 if (mdt_fids_different_target(info, &pfid,
1982 mdt_object_fid(child)))
1989 int mdt_pack_encctx_in_reply(struct mdt_thread_info *info,
1990 struct mdt_object *child)
1992 struct lu_buf *buffer;
1993 struct mdt_body *repbody;
1994 struct req_capsule *pill = info->mti_pill;
1995 struct obd_export *exp = mdt_info_req(info)->rq_export;
1998 if (!exp_connect_encrypt(exp))
2001 if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
2002 req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0) {
2003 struct lu_attr la = { 0 };
2004 struct dt_object *dt = mdt_obj2dt(child);
2006 if (dt && dt->do_ops && dt->do_ops->do_attr_get)
2007 dt_attr_get(info->mti_env, mdt_obj2dt(child), &la);
2009 if (la.la_valid & LA_FLAGS && la.la_flags & LUSTRE_ENCRYPT_FL) {
2010 buffer = &info->mti_buf;
2012 /* fill reply buffer with encryption context now */
2014 req_capsule_get_size(pill, &RMF_FILE_ENCCTX,
2017 req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
2018 rc = mo_xattr_get(info->mti_env,
2019 mdt_object_child(child),
2021 LL_XATTR_NAME_ENCRYPTION_CONTEXT);
2024 "found encryption ctx of size %d for "DFID"\n",
2025 rc, PFID(mdt_object_fid(child)));
2027 repbody = req_capsule_server_get(pill,
2029 repbody->mbo_valid |= OBD_MD_ENCCTX;
2030 if (rc < buffer->lb_len)
2031 req_capsule_shrink(pill,
2032 &RMF_FILE_ENCCTX, rc,
2037 "encryption ctx not found for "DFID": rc = %d\n",
2038 PFID(mdt_object_fid(child)), rc);
2039 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,
2041 /* handling -ENOENT is important because it may
2042 * change object state in DNE env dropping
2043 * LOHA_EXISTS flag, it is important to return
2044 * that to the caller.
2045 * Check LU-13115 for details.
2051 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,