Whamcloud - gitweb
998d581bd6a3d77b3b3459f6ed81e25b1d574a41
[fs/lustre-release.git] / lustre / mdt / mdt_lib.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9  *
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).
15  *
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
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2011, 2016, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  *
32  * lustre/mdt/mdt_lib.c
33  *
34  * Lustre Metadata Target (mdt) request unpacking helper.
35  *
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>
43  */
44
45 #define DEBUG_SUBSYSTEM S_MDS
46
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
50 #endif
51 #include "mdt_internal.h"
52 #include <uapi/linux/lnet/nidstr.h>
53 #include <lustre_nodemap.h>
54
55 typedef enum ucred_init_type {
56         NONE_INIT       = 0,
57         BODY_INIT       = 1,
58         REC_INIT        = 2
59 } ucred_init_type_t;
60
61 static __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
62 {
63         return (__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32);
64 }
65
66 void mdt_exit_ucred(struct mdt_thread_info *info)
67 {
68         struct lu_ucred   *uc  = mdt_ucred(info);
69         struct mdt_device *mdt = info->mti_mdt;
70
71         LASSERT(uc != NULL);
72         if (uc->uc_valid != UCRED_INIT) {
73                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
74                 if (uc->uc_ginfo) {
75                         put_group_info(uc->uc_ginfo);
76                         uc->uc_ginfo = NULL;
77                 }
78                 if (uc->uc_identity) {
79                         mdt_identity_put(mdt->mdt_identity_cache,
80                                          uc->uc_identity);
81                         uc->uc_identity = NULL;
82                 }
83                 uc->uc_valid = UCRED_INIT;
84         }
85 }
86
87 static int match_nosquash_list(struct rw_semaphore *sem,
88                                struct list_head *nidlist,
89                                lnet_nid_t peernid)
90 {
91         int rc;
92         ENTRY;
93         down_read(sem);
94         rc = cfs_match_nid(peernid, nidlist);
95         up_read(sem);
96         RETURN(rc);
97 }
98
99 /* root_squash for inter-MDS operations */
100 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
101 {
102         struct lu_ucred *ucred = mdt_ucred(info);
103         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
104         ENTRY;
105
106         LASSERT(ucred != NULL);
107         if (!squash->rsi_uid || ucred->uc_fsuid)
108                 RETURN(0);
109
110         if (match_nosquash_list(&squash->rsi_sem,
111                                 &squash->rsi_nosquash_nids,
112                                 peernid)) {
113                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
114                        libcfs_nid2str(peernid));
115                 RETURN(0);
116         }
117
118         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
119                libcfs_nid2str(peernid),
120                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
121                squash->rsi_uid, squash->rsi_gid, 0);
122
123         ucred->uc_fsuid = squash->rsi_uid;
124         ucred->uc_fsgid = squash->rsi_gid;
125         ucred->uc_cap = 0;
126         ucred->uc_suppgids[0] = -1;
127         ucred->uc_suppgids[1] = -1;
128
129         RETURN(0);
130 }
131
132 static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
133 {
134         struct ptlrpc_request   *req = mdt_info_req(info);
135         const char              *jobid = mdt_req_get_jobid(req);
136
137         /* set jobid if specified. */
138         if (jobid)
139                 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
140         else
141                 uc->uc_jobid[0] = '\0';
142 }
143
144 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
145                           void *buf, bool drop_fs_cap)
146 {
147         struct ptlrpc_request   *req = mdt_info_req(info);
148         struct mdt_device       *mdt = info->mti_mdt;
149         struct ptlrpc_user_desc *pud = req->rq_user_desc;
150         struct lu_ucred         *ucred = mdt_ucred(info);
151         lnet_nid_t               peernid = req->rq_peer.nid;
152         __u32                    perm = 0;
153         int                      setuid;
154         int                      setgid;
155         int                      rc = 0;
156
157         ENTRY;
158
159         LASSERT(req->rq_auth_gss);
160         LASSERT(!req->rq_auth_usr_mdt);
161         LASSERT(req->rq_user_desc);
162         LASSERT(ucred != NULL);
163
164         ucred->uc_valid = UCRED_INVALID;
165
166         ucred->uc_o_uid   = pud->pud_uid;
167         ucred->uc_o_gid   = pud->pud_gid;
168         ucred->uc_o_fsuid = pud->pud_fsuid;
169         ucred->uc_o_fsgid = pud->pud_fsgid;
170
171         if (type == BODY_INIT) {
172                 struct mdt_body *body = (struct mdt_body *)buf;
173
174                 ucred->uc_suppgids[0] = body->mbo_suppgid;
175                 ucred->uc_suppgids[1] = -1;
176         }
177
178         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
179             req->rq_auth_uid != pud->pud_uid) {
180                 CDEBUG(D_SEC, "local client %s: auth uid %u "
181                        "while client claims %u:%u/%u:%u\n",
182                        libcfs_nid2str(peernid), req->rq_auth_uid,
183                        pud->pud_uid, pud->pud_gid,
184                        pud->pud_fsuid, pud->pud_fsgid);
185                 RETURN(-EACCES);
186         }
187
188         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
189                 ucred->uc_identity = NULL;
190                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
191         } else {
192                 struct md_identity *identity;
193
194                 identity = mdt_identity_get(mdt->mdt_identity_cache,
195                                             pud->pud_uid);
196                 if (IS_ERR(identity)) {
197                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
198                                 ucred->uc_identity = NULL;
199                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
200                                        CFS_SETGRP_PERM;
201                         } else {
202                                 CDEBUG(D_SEC,
203                                        "Deny access without identity: uid %u\n",
204                                        pud->pud_uid);
205                                 RETURN(-EACCES);
206                         }
207                 } else {
208                         ucred->uc_identity = identity;
209                         perm = mdt_identity_get_perm(ucred->uc_identity,
210                                                      peernid);
211                 }
212         }
213
214         /* find out the setuid/setgid attempt */
215         setuid = (pud->pud_uid != pud->pud_fsuid);
216         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
217                   (ucred->uc_identity &&
218                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
219
220         /* check permission of setuid */
221         if (setuid && !(perm & CFS_SETUID_PERM)) {
222                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
223                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
224                 GOTO(out, rc = -EACCES);
225         }
226
227         /* check permission of setgid */
228         if (setgid && !(perm & CFS_SETGID_PERM)) {
229                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
230                        "from %s\n", pud->pud_uid, pud->pud_gid,
231                        pud->pud_fsuid, pud->pud_fsgid,
232                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
233                 GOTO(out, rc = -EACCES);
234         }
235
236         if (perm & CFS_SETGRP_PERM) {
237                 if (pud->pud_ngroups) {
238                         /* setgroups for local client */
239                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
240                         if (!ucred->uc_ginfo) {
241                                 CERROR("failed to alloc %d groups\n",
242                                        pud->pud_ngroups);
243                                 GOTO(out, rc = -ENOMEM);
244                         }
245
246                         lustre_groups_from_list(ucred->uc_ginfo,
247                                                 pud->pud_groups);
248                         lustre_groups_sort(ucred->uc_ginfo);
249                 } else {
250                         ucred->uc_ginfo = NULL;
251                 }
252         } else {
253                 ucred->uc_suppgids[0] = -1;
254                 ucred->uc_suppgids[1] = -1;
255                 ucred->uc_ginfo = NULL;
256         }
257
258         ucred->uc_uid   = pud->pud_uid;
259         ucred->uc_gid   = pud->pud_gid;
260         ucred->uc_fsuid = pud->pud_fsuid;
261         ucred->uc_fsgid = pud->pud_fsgid;
262
263         /* process root_squash here. */
264         mdt_root_squash(info, peernid);
265
266         /* remove fs privilege for non-root user. */
267         if (ucred->uc_fsuid && drop_fs_cap)
268                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
269         else
270                 ucred->uc_cap = pud->pud_cap;
271         ucred->uc_valid = UCRED_NEW;
272         ucred_set_jobid(info, ucred);
273
274         EXIT;
275
276 out:
277         if (rc) {
278                 if (ucred->uc_ginfo) {
279                         put_group_info(ucred->uc_ginfo);
280                         ucred->uc_ginfo = NULL;
281                 }
282                 if (ucred->uc_identity) {
283                         mdt_identity_put(mdt->mdt_identity_cache,
284                                          ucred->uc_identity);
285                         ucred->uc_identity = NULL;
286                 }
287         }
288
289         return rc;
290 }
291
292 /**
293  * Check whether allow the client to set supplementary group IDs or not.
294  *
295  * \param[in] info      pointer to the thread context
296  * \param[in] uc        pointer to the RPC user descriptor
297  *
298  * \retval              true if allow to set supplementary group IDs
299  * \retval              false for other cases
300  */
301 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
302 {
303         __u32 perm;
304
305         /* 1. If identity_upcall is disabled,
306          *    permit local client to do anything. */
307         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
308                 return true;
309
310         /* 2. If fail to get related identities, then forbid any client to
311          *    set supplementary group IDs. */
312         if (uc->uc_identity == NULL)
313                 return false;
314
315         /* 3. Check the permission in the identities. */
316         perm = mdt_identity_get_perm(uc->uc_identity,
317                                      mdt_info_req(info)->rq_peer.nid);
318         if (perm & CFS_SETGRP_PERM)
319                 return true;
320
321         return false;
322 }
323
324 int mdt_check_ucred(struct mdt_thread_info *info)
325 {
326         struct ptlrpc_request   *req = mdt_info_req(info);
327         struct mdt_device       *mdt = info->mti_mdt;
328         struct ptlrpc_user_desc *pud = req->rq_user_desc;
329         struct lu_ucred         *ucred = mdt_ucred(info);
330         struct md_identity      *identity = NULL;
331         lnet_nid_t               peernid = req->rq_peer.nid;
332         __u32                    perm = 0;
333         int                      setuid;
334         int                      setgid;
335         int                      rc = 0;
336
337         ENTRY;
338
339         LASSERT(ucred != NULL);
340         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
341                 RETURN(0);
342
343         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
344                 RETURN(0);
345
346         /* sanity check: if we use strong authentication, we expect the
347          * uid which client claimed is true */
348         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
349             req->rq_auth_uid != pud->pud_uid) {
350                 CDEBUG(D_SEC, "local client %s: auth uid %u "
351                        "while client claims %u:%u/%u:%u\n",
352                        libcfs_nid2str(peernid), req->rq_auth_uid,
353                        pud->pud_uid, pud->pud_gid,
354                        pud->pud_fsuid, pud->pud_fsgid);
355                 RETURN(-EACCES);
356         }
357
358         if (is_identity_get_disabled(mdt->mdt_identity_cache))
359                 RETURN(0);
360
361         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
362         if (IS_ERR(identity)) {
363                 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
364                         RETURN(0);
365                 } else {
366                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
367                                pud->pud_uid);
368                         RETURN(-EACCES);
369                 }
370         }
371
372         perm = mdt_identity_get_perm(identity, peernid);
373         /* find out the setuid/setgid attempt */
374         setuid = (pud->pud_uid != pud->pud_fsuid);
375         setgid = (pud->pud_gid != pud->pud_fsgid ||
376                   pud->pud_gid != identity->mi_gid);
377
378         /* check permission of setuid */
379         if (setuid && !(perm & CFS_SETUID_PERM)) {
380                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
381                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
382                 GOTO(out, rc = -EACCES);
383         }
384
385         /* check permission of setgid */
386         if (setgid && !(perm & CFS_SETGID_PERM)) {
387                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
388                        "from %s\n", pud->pud_uid, pud->pud_gid,
389                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
390                        libcfs_nid2str(peernid));
391                 GOTO(out, rc = -EACCES);
392         }
393
394         EXIT;
395
396 out:
397         mdt_identity_put(mdt->mdt_identity_cache, identity);
398         return rc;
399 }
400
401 static int old_init_ucred_common(struct mdt_thread_info *info,
402                                  struct lu_nodemap *nodemap,
403                                  bool drop_fs_cap)
404 {
405         struct lu_ucred         *uc = mdt_ucred(info);
406         struct mdt_device       *mdt = info->mti_mdt;
407         struct md_identity      *identity = NULL;
408
409         if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
410                 /* deny access before we get identity ref */
411                 if (nodemap->nmf_deny_unknown)
412                         RETURN(-EACCES);
413
414                 uc->uc_fsuid = nodemap->nm_squash_uid;
415                 uc->uc_fsgid = nodemap->nm_squash_gid;
416                 uc->uc_cap = 0;
417                 uc->uc_suppgids[0] = -1;
418                 uc->uc_suppgids[1] = -1;
419         }
420
421         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
422                 identity = mdt_identity_get(mdt->mdt_identity_cache,
423                                             uc->uc_fsuid);
424                 if (IS_ERR(identity)) {
425                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
426                                      uc->uc_cap & CFS_CAP_FS_MASK)) {
427                                 identity = NULL;
428                         } else {
429                                 CDEBUG(D_SEC, "Deny access without identity: "
430                                        "uid %u\n", uc->uc_fsuid);
431                                 RETURN(-EACCES);
432                         }
433                 }
434         }
435         uc->uc_identity = identity;
436
437         /* process root_squash here. */
438         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
439
440         /* remove fs privilege for non-root user. */
441         if (uc->uc_fsuid && drop_fs_cap)
442                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
443         uc->uc_valid = UCRED_OLD;
444         ucred_set_jobid(info, uc);
445
446         EXIT;
447
448         return 0;
449 }
450
451 static int old_init_ucred(struct mdt_thread_info *info,
452                           struct mdt_body *body, bool drop_fs_cap)
453 {
454         struct lu_ucred *uc = mdt_ucred(info);
455         struct lu_nodemap *nodemap;
456         int rc;
457         ENTRY;
458
459         nodemap = nodemap_get_from_exp(info->mti_exp);
460         if (IS_ERR(nodemap))
461                 RETURN(PTR_ERR(nodemap));
462
463         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
464                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
465         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
466                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
467         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
468                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
469         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
470                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
471
472         LASSERT(uc != NULL);
473         uc->uc_valid = UCRED_INVALID;
474         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
475         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
476         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
477         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
478         uc->uc_suppgids[0] = body->mbo_suppgid;
479         uc->uc_suppgids[1] = -1;
480         uc->uc_ginfo = NULL;
481         uc->uc_cap = body->mbo_capability;
482
483         rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
484         nodemap_putref(nodemap);
485
486         RETURN(rc);
487 }
488
489 static int old_init_ucred_reint(struct mdt_thread_info *info)
490 {
491         struct lu_ucred *uc = mdt_ucred(info);
492         struct lu_nodemap *nodemap;
493         int rc;
494         ENTRY;
495
496         nodemap = nodemap_get_from_exp(info->mti_exp);
497         if (IS_ERR(nodemap))
498                 RETURN(PTR_ERR(nodemap));
499
500         LASSERT(uc != NULL);
501
502         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
503                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
504         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
505                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
506
507         uc->uc_valid = UCRED_INVALID;
508         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
509         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
510         uc->uc_ginfo = NULL;
511
512         rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */
513         nodemap_putref(nodemap);
514
515         RETURN(rc);
516 }
517
518 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
519                                    struct mdt_body *body,
520                                    bool drop_fs_cap)
521 {
522         struct ptlrpc_request   *req = mdt_info_req(info);
523         struct lu_ucred         *uc  = mdt_ucred(info);
524
525         LASSERT(uc != NULL);
526         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
527                 return 0;
528
529         mdt_exit_ucred(info);
530
531         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
532                 return old_init_ucred(info, body, drop_fs_cap);
533         else
534                 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
535 }
536
537 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
538 {
539         return __mdt_init_ucred(info, body, true);
540 }
541
542 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
543  * doesn't set correct fsuid explicitely, but raise capability to allow
544  * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
545  * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
546  * will fail permission check. */
547 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
548                                   struct mdt_body *body)
549 {
550         return __mdt_init_ucred(info, body, false);
551 }
552
553 int mdt_init_ucred_reint(struct mdt_thread_info *info)
554 {
555         struct ptlrpc_request *req = mdt_info_req(info);
556         struct lu_ucred       *uc  = mdt_ucred(info);
557         struct md_attr        *ma  = &info->mti_attr;
558
559         LASSERT(uc != NULL);
560         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
561                 return 0;
562
563         /* LU-5564: for normal close request, skip permission check */
564         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
565             !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
566                 uc->uc_cap |= CFS_CAP_FS_MASK;
567
568         mdt_exit_ucred(info);
569
570         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
571                 return old_init_ucred_reint(info);
572         else
573                 return new_init_ucred(info, REC_INIT, NULL, true);
574 }
575
576 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
577 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
578 {
579         const struct lov_ost_data_v1    *lod;
580         int                              i;
581         __u16                            count;
582
583         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
584                 return;
585
586         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
587
588         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
589                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
590                le32_to_cpu(lmm->lmm_pattern));
591         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
592                le32_to_cpu(lmm->lmm_stripe_size), count);
593
594         /* If it's a directory or a released file, then there are
595          * no actual objects to print, so bail out. */
596         if (valid & OBD_MD_FLDIREA ||
597             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
598                 return;
599
600         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
601         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
602                 struct ost_id oi;
603
604                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
605                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
606                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
607         }
608 }
609
610 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
611 {
612         const struct lmv_mds_md_v1 *lmm1;
613         int                        i;
614
615         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
616                 return;
617
618         lmm1 = &lmv->lmv_md_v1;
619         CDEBUG(level,
620                "magic 0x%08X, master %#X stripe_count %#x hash_type %#x\n",
621                le32_to_cpu(lmm1->lmv_magic),
622                le32_to_cpu(lmm1->lmv_master_mdt_index),
623                le32_to_cpu(lmm1->lmv_stripe_count),
624                le32_to_cpu(lmm1->lmv_hash_type));
625
626         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
627                 return;
628
629         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
630                 struct lu_fid fid;
631
632                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
633                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
634         }
635 }
636
637 /* Shrink and/or grow reply buffers */
638 int mdt_fix_reply(struct mdt_thread_info *info)
639 {
640         struct req_capsule *pill = info->mti_pill;
641         struct mdt_body    *body;
642         int                md_size, md_packed = 0;
643         int                acl_size;
644         int                rc = 0;
645         ENTRY;
646
647         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
648         LASSERT(body != NULL);
649
650         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
651                                OBD_MD_LINKNAME))
652                 md_size = body->mbo_eadatasize;
653         else
654                 md_size = 0;
655
656         acl_size = body->mbo_aclsize;
657
658         /* this replay - not send info to client */
659         if (info->mti_spec.no_create) {
660                 md_size = 0;
661                 acl_size = 0;
662         }
663
664         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
665                md_size, acl_size);
666 /*
667             &RMF_MDT_BODY,
668             &RMF_MDT_MD,
669             &RMF_ACL, or &RMF_LOGCOOKIES
670 (optional)  &RMF_CAPA1,
671 (optional)  &RMF_CAPA2,
672 (optional)  something else
673 */
674
675         /* MDT_MD buffer may be bigger than packed value, let's shrink all
676          * buffers before growing it */
677         if (info->mti_big_lmm_used) {
678                 /* big_lmm buffer may be used even without packing the result
679                  * into reply, just for internal server needs */
680                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
681                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
682                                                          RCL_SERVER);
683
684                 /* free big lmm if md_size is not needed */
685                 if (md_size == 0 || md_packed == 0) {
686                         info->mti_big_lmm_used = 0;
687                 } else {
688                         /* buffer must be allocated separately */
689                         LASSERT(info->mti_attr.ma_lmm !=
690                                 req_capsule_server_get(pill, &RMF_MDT_MD));
691                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
692                 }
693         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
694                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
695         }
696
697         if (info->mti_big_acl_used) {
698                 if (acl_size == 0)
699                         info->mti_big_acl_used = 0;
700                 else
701                         req_capsule_shrink(pill, &RMF_ACL, 0, RCL_SERVER);
702         } else if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER)) {
703                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
704         } else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) {
705                 req_capsule_shrink(pill, &RMF_LOGCOOKIES, acl_size, RCL_SERVER);
706         }
707
708         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
709             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
710                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
711
712         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
713             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
714                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
715
716         /*
717          * Some more field should be shrinked if needed.
718          * This should be done by those who added fields to reply message.
719          */
720
721         /* Grow MD buffer if needed finally */
722         if (info->mti_big_lmm_used) {
723                 void *lmm;
724
725                 LASSERT(md_size > md_packed);
726                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
727                        md_size - md_packed);
728                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
729                 if (rc) {
730                         /* we can't answer with proper LOV EA, drop flags,
731                          * the rc is also returned so this request is
732                          * considered as failed */
733                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
734                         /* don't return transno along with error */
735                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
736                 } else {
737                         /* now we need to pack right LOV/LMV EA */
738                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
739                         if (info->mti_attr.ma_valid & MA_LOV) {
740                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
741                                                              RCL_SERVER) ==
742                                                 info->mti_attr.ma_lmm_size);
743                                 memcpy(lmm, info->mti_attr.ma_lmm,
744                                        info->mti_attr.ma_lmm_size);
745                         } else if (info->mti_attr.ma_valid & MA_LMV) {
746                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
747                                                              RCL_SERVER) ==
748                                                 info->mti_attr.ma_lmv_size);
749                                 memcpy(lmm, info->mti_attr.ma_lmv,
750                                        info->mti_attr.ma_lmv_size);
751                         }
752                 }
753
754                 /* update mdt_max_mdsize so clients will be aware about that */
755                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
756                         info->mti_mdt->mdt_max_mdsize =
757                                                 info->mti_attr.ma_lmm_size;
758                 info->mti_big_lmm_used = 0;
759         }
760
761         if (info->mti_big_acl_used) {
762                 CDEBUG(D_INFO, "Enlarge reply ACL buffer to %d bytes\n",
763                        acl_size);
764
765                 rc = req_capsule_server_grow(pill, &RMF_ACL, acl_size);
766                 if (rc) {
767                         body->mbo_valid &= ~OBD_MD_FLACL;
768                 } else {
769                         void *acl = req_capsule_server_get(pill, &RMF_ACL);
770
771                         memcpy(acl, info->mti_big_acl, acl_size);
772                 }
773
774                 info->mti_big_acl_used = 0;
775         }
776
777         RETURN(rc);
778 }
779
780
781 /* if object is dying, pack the lov/llog data,
782  * parameter info->mti_attr should be valid at this point!
783  * Also implements RAoLU policy */
784 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
785                            struct md_attr *ma)
786 {
787         struct mdt_body *repbody = NULL;
788         const struct lu_attr *la = &ma->ma_attr;
789         struct coordinator *cdt = &info->mti_mdt->mdt_coordinator;
790         int rc;
791         __u64 need = 0;
792         struct hsm_action_item hai = {
793                 .hai_len = sizeof(hai),
794                 .hai_action = HSMA_REMOVE,
795                 .hai_extent.length = -1,
796                 .hai_cookie = 0,
797                 .hai_gid = 0,
798         };
799         __u64 compound_id;
800         int archive_id;
801
802         ENTRY;
803
804         if (mdt_info_req(info) != NULL) {
805                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
806                 LASSERT(repbody != NULL);
807         } else {
808                 CDEBUG(D_INFO, "not running in a request/reply context\n");
809         }
810
811         if ((ma->ma_valid & MA_INODE) && repbody != NULL)
812                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
813
814         if (ma->ma_valid & MA_LOV) {
815                 CERROR("No need in LOV EA upon unlink\n");
816                 dump_stack();
817         }
818         if (repbody != NULL)
819                 repbody->mbo_eadatasize = 0;
820
821         /* Only check unlinked and archived if RAoLU and upon last close */
822         if (!cdt->cdt_remove_archive_on_last_unlink ||
823             atomic_read(&mo->mot_open_count) != 0)
824                 RETURN(0);
825
826         /* mdt_attr_get_complex will clear ma_valid, so check here first */
827         if ((ma->ma_valid & MA_INODE) && (ma->ma_attr.la_nlink != 0))
828                 RETURN(0);
829
830         if ((ma->ma_valid & MA_HSM) && (!(ma->ma_hsm.mh_flags & HS_EXISTS)))
831                 RETURN(0);
832
833         need |= (MA_INODE | MA_HSM) & ~ma->ma_valid;
834         if (need != 0) {
835                 /* ma->ma_valid is missing either MA_INODE, MA_HSM, or both,
836                  * try setting them */
837                 ma->ma_need |= need;
838                 rc = mdt_attr_get_complex(info, mo, ma);
839                 if (rc) {
840                         CERROR("%s: unable to fetch missing attributes of"
841                                DFID": rc=%d\n", mdt_obd_name(info->mti_mdt),
842                                PFID(mdt_object_fid(mo)), rc);
843                         RETURN(0);
844                 }
845
846                 if (need & MA_INODE) {
847                         if (ma->ma_valid & MA_INODE) {
848                                 if (ma->ma_attr.la_nlink != 0)
849                                         RETURN(0);
850                         } else {
851                                 RETURN(0);
852                         }
853                 }
854
855                 if (need & MA_HSM) {
856                         if (ma->ma_valid & MA_HSM) {
857                                 if (!(ma->ma_hsm.mh_flags & HS_EXISTS))
858                                         RETURN(0);
859                         } else {
860                                 RETURN(0);
861                         }
862                 }
863         }
864
865         /* RAoLU policy is active, last close on file has occured,
866          * file is unlinked, file is archived, so create remove request
867          * for copytool!
868          * If CDT is not running, requests will be logged for later. */
869         compound_id = atomic_inc_return(&cdt->cdt_compound_id);
870         if (ma->ma_hsm.mh_arch_id != 0)
871                 archive_id = ma->ma_hsm.mh_arch_id;
872         else
873                 archive_id = cdt->cdt_default_archive_id;
874
875         hai.hai_fid = *mdt_object_fid(mo);
876
877         rc = mdt_agent_record_add(info->mti_env, info->mti_mdt,
878                                   compound_id, archive_id, 0, &hai);
879         if (rc)
880                 CERROR("%s: unable to add HSM remove request for "DFID
881                        ": rc=%d\n", mdt_obd_name(info->mti_mdt),
882                        PFID(mdt_object_fid(mo)), rc);
883
884         RETURN(0);
885 }
886
887 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
888                                   struct md_attr *ma)
889 {
890         __u64 out;
891
892         out = 0;
893         if (in & MDS_ATTR_MODE)
894                 out |= LA_MODE;
895         if (in & MDS_ATTR_UID)
896                 out |= LA_UID;
897         if (in & MDS_ATTR_GID)
898                 out |= LA_GID;
899         if (in & MDS_ATTR_SIZE)
900                 out |= LA_SIZE;
901         if (in & MDS_ATTR_BLOCKS)
902                 out |= LA_BLOCKS;
903         if (in & MDS_ATTR_ATIME_SET)
904                 out |= LA_ATIME;
905         if (in & MDS_ATTR_CTIME_SET)
906                 out |= LA_CTIME;
907         if (in & MDS_ATTR_MTIME_SET)
908                 out |= LA_MTIME;
909         if (in & MDS_ATTR_ATTR_FLAG)
910                 out |= LA_FLAGS;
911         if (in & MDS_ATTR_KILL_SUID)
912                 out |= LA_KILL_SUID;
913         if (in & MDS_ATTR_KILL_SGID)
914                 out |= LA_KILL_SGID;
915         if (in & MDS_ATTR_PROJID)
916                 out |= LA_PROJID;
917
918         if (in & MDS_ATTR_FROM_OPEN)
919                 rr->rr_flags |= MRF_OPEN_TRUNC;
920         if (in & MDS_OPEN_OWNEROVERRIDE)
921                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
922         if (in & MDS_ATTR_FORCE)
923                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
924
925         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | MDS_ATTR_PROJID |
926                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
927                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
928                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
929                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
930                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
931         if (in != 0)
932                 CERROR("Unknown attr bits: %#llx\n", in);
933         return out;
934 }
935
936 /* unpacking */
937
938 int mdt_name_unpack(struct req_capsule *pill,
939                     const struct req_msg_field *field,
940                     struct lu_name *ln,
941                     enum mdt_name_flags flags)
942 {
943         ln->ln_name = req_capsule_client_get(pill, field);
944         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
945
946         if (!lu_name_is_valid(ln)) {
947                 ln->ln_name = NULL;
948                 ln->ln_namelen = 0;
949
950                 return -EPROTO;
951         }
952
953         if ((flags & MNF_FIX_ANON) &&
954             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
955                 /* Newer (3.x) kernels use a name of "/" for the
956                  * "anonymous" disconnected dentries from NFS
957                  * filehandle conversion. See d_obtain_alias(). */
958                 ln->ln_name = NULL;
959                 ln->ln_namelen = 0;
960         }
961
962         return 0;
963 }
964
965 static int mdt_file_secctx_unpack(struct req_capsule *pill,
966                                   const char **secctx_name,
967                                   void **secctx, size_t *secctx_size)
968 {
969         const char *name;
970         size_t name_size;
971
972         *secctx_name = NULL;
973         *secctx = NULL;
974         *secctx_size = 0;
975
976         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
977             !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
978                 return 0;
979
980         name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
981                                          RCL_CLIENT);
982         if (name_size == 0)
983                 return 0;
984
985         name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
986         if (strnlen(name, name_size) != name_size - 1)
987                 return -EPROTO;
988
989         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
990             !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
991                 return -EPROTO;
992
993         *secctx_name = name;
994         *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
995         *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
996
997         return 0;
998 }
999
1000 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
1001 {
1002         struct lu_ucred         *uc = mdt_ucred(info);
1003         struct md_attr          *ma = &info->mti_attr;
1004         struct lu_attr          *la = &ma->ma_attr;
1005         struct req_capsule      *pill = info->mti_pill;
1006         struct mdt_reint_record *rr = &info->mti_rr;
1007         struct mdt_rec_setattr  *rec;
1008         struct lu_nodemap       *nodemap;
1009         ENTRY;
1010
1011         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
1012         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1013         if (rec == NULL)
1014                 RETURN(-EFAULT);
1015
1016         /* This prior initialization is needed for old_init_ucred_reint() */
1017         uc->uc_fsuid = rec->sa_fsuid;
1018         uc->uc_fsgid = rec->sa_fsgid;
1019         uc->uc_cap   = rec->sa_cap;
1020         uc->uc_suppgids[0] = rec->sa_suppgid;
1021         uc->uc_suppgids[1] = -1;
1022
1023         rr->rr_fid1 = &rec->sa_fid;
1024         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
1025         la->la_mode  = rec->sa_mode;
1026         la->la_flags = rec->sa_attr_flags;
1027
1028         nodemap = nodemap_get_from_exp(info->mti_exp);
1029         if (IS_ERR(nodemap))
1030                 RETURN(PTR_ERR(nodemap));
1031
1032         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
1033                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1034         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
1035                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1036         la->la_projid = rec->sa_projid;
1037         nodemap_putref(nodemap);
1038
1039         la->la_size  = rec->sa_size;
1040         la->la_blocks = rec->sa_blocks;
1041         la->la_ctime = rec->sa_ctime;
1042         la->la_atime = rec->sa_atime;
1043         la->la_mtime = rec->sa_mtime;
1044         ma->ma_valid = MA_INODE;
1045
1046         if (rec->sa_bias & MDS_DATA_MODIFIED)
1047                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
1048         else
1049                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
1050
1051         if (rec->sa_bias & MDS_HSM_RELEASE)
1052                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
1053         else
1054                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
1055
1056         if (rec->sa_bias & MDS_CLOSE_LAYOUT_SWAP)
1057                 ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP;
1058         else
1059                 ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP;
1060
1061         RETURN(0);
1062 }
1063
1064 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
1065 {
1066         struct req_capsule *pill = info->mti_pill;
1067         struct mdt_ioepoch *ioepoch;
1068         ENTRY;
1069
1070         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1071                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1072         else
1073                 ioepoch = NULL;
1074
1075         if (ioepoch == NULL)
1076                 RETURN(-EPROTO);
1077
1078         info->mti_close_handle = ioepoch->mio_handle;
1079
1080         RETURN(0);
1081 }
1082
1083 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1084         struct req_capsule      *pill = info->mti_pill;
1085
1086         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1087                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1088                 if (info->mti_dlm_req == NULL)
1089                         RETURN(-EFAULT);
1090         }
1091
1092         RETURN(0);
1093 }
1094
1095 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1096 {
1097         struct mdt_reint_record *rr = &info->mti_rr;
1098         struct md_attr          *ma = &info->mti_attr;
1099         struct req_capsule      *pill = info->mti_pill;
1100         int rc;
1101         ENTRY;
1102
1103         rc = mdt_setattr_unpack_rec(info);
1104         if (rc)
1105                 RETURN(rc);
1106
1107         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1108                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1109                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1110                                                         RCL_CLIENT);
1111                 if (rr->rr_eadatalen > 0) {
1112                         const struct lmv_user_md        *lum;
1113
1114                         lum = rr->rr_eadata;
1115                         /* Sigh ma_valid(from req) does not indicate whether
1116                          * it will set LOV/LMV EA, so we have to check magic */
1117                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1118                                 ma->ma_valid |= MA_LMV;
1119                                 ma->ma_lmv = (void *)rr->rr_eadata;
1120                                 ma->ma_lmv_size = rr->rr_eadatalen;
1121                         } else {
1122                                 ma->ma_valid |= MA_LOV;
1123                                 ma->ma_lmm = (void *)rr->rr_eadata;
1124                                 ma->ma_lmm_size = rr->rr_eadatalen;
1125                         }
1126                 }
1127         }
1128
1129         rc = mdt_dlmreq_unpack(info);
1130         RETURN(rc);
1131 }
1132
1133 static int mdt_intent_close_unpack(struct mdt_thread_info *info)
1134 {
1135         struct md_attr          *ma = &info->mti_attr;
1136         struct req_capsule      *pill = info->mti_pill;
1137         ENTRY;
1138
1139         if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
1140                 RETURN(0);
1141
1142         req_capsule_extend(pill, &RQF_MDS_INTENT_CLOSE);
1143
1144         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1145             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1146                 RETURN(-EFAULT);
1147
1148         RETURN(0);
1149 }
1150
1151 int mdt_close_unpack(struct mdt_thread_info *info)
1152 {
1153         int rc;
1154         ENTRY;
1155
1156         rc = mdt_close_handle_unpack(info);
1157         if (rc)
1158                 RETURN(rc);
1159
1160         rc = mdt_setattr_unpack_rec(info);
1161         if (rc)
1162                 RETURN(rc);
1163
1164         rc = mdt_intent_close_unpack(info);
1165         if (rc)
1166                 RETURN(rc);
1167
1168         RETURN(mdt_init_ucred_reint(info));
1169 }
1170
1171 static int mdt_create_unpack(struct mdt_thread_info *info)
1172 {
1173         struct lu_ucred         *uc  = mdt_ucred(info);
1174         struct mdt_rec_create   *rec;
1175         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1176         struct mdt_reint_record *rr = &info->mti_rr;
1177         struct req_capsule      *pill = info->mti_pill;
1178         struct md_op_spec       *sp = &info->mti_spec;
1179         int rc;
1180         ENTRY;
1181
1182         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1183         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1184         if (rec == NULL)
1185                 RETURN(-EFAULT);
1186
1187         /* This prior initialization is needed for old_init_ucred_reint() */
1188         uc->uc_fsuid = rec->cr_fsuid;
1189         uc->uc_fsgid = rec->cr_fsgid;
1190         uc->uc_cap   = rec->cr_cap;
1191         uc->uc_suppgids[0] = rec->cr_suppgid1;
1192         uc->uc_suppgids[1] = -1;
1193         uc->uc_umask = rec->cr_umask;
1194
1195         rr->rr_fid1 = &rec->cr_fid1;
1196         rr->rr_fid2 = &rec->cr_fid2;
1197         attr->la_mode = rec->cr_mode;
1198         attr->la_rdev  = rec->cr_rdev;
1199         attr->la_uid   = rec->cr_fsuid;
1200         attr->la_gid   = rec->cr_fsgid;
1201         attr->la_ctime = rec->cr_time;
1202         attr->la_mtime = rec->cr_time;
1203         attr->la_atime = rec->cr_time;
1204         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1205                          LA_CTIME | LA_MTIME | LA_ATIME;
1206         memset(&sp->u, 0, sizeof(sp->u));
1207         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1208
1209         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1210         if (rc < 0)
1211                 RETURN(rc);
1212
1213         if (S_ISLNK(attr->la_mode)) {
1214                 const char *tgt = NULL;
1215
1216                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1217                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1218                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1219                         sp->u.sp_symname = tgt;
1220                 }
1221                 if (tgt == NULL)
1222                         RETURN(-EFAULT);
1223         } else {
1224                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1225                 if (S_ISDIR(attr->la_mode) &&
1226                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1227                         sp->u.sp_ea.eadata =
1228                                 req_capsule_client_get(pill, &RMF_EADATA);
1229                         sp->u.sp_ea.eadatalen =
1230                                 req_capsule_get_size(pill, &RMF_EADATA,
1231                                                      RCL_CLIENT);
1232                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1233                 }
1234         }
1235
1236         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1237                                     &sp->sp_cr_file_secctx,
1238                                     &sp->sp_cr_file_secctx_size);
1239         if (rc < 0)
1240                 RETURN(rc);
1241
1242         rc = mdt_dlmreq_unpack(info);
1243         RETURN(rc);
1244 }
1245
1246 static int mdt_link_unpack(struct mdt_thread_info *info)
1247 {
1248         struct lu_ucred         *uc  = mdt_ucred(info);
1249         struct mdt_rec_link     *rec;
1250         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1251         struct mdt_reint_record *rr = &info->mti_rr;
1252         struct req_capsule      *pill = info->mti_pill;
1253         int rc;
1254         ENTRY;
1255
1256         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1257         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1258         if (rec == NULL)
1259                 RETURN(-EFAULT);
1260
1261         /* This prior initialization is needed for old_init_ucred_reint() */
1262         uc->uc_fsuid = rec->lk_fsuid;
1263         uc->uc_fsgid = rec->lk_fsgid;
1264         uc->uc_cap   = rec->lk_cap;
1265         uc->uc_suppgids[0] = rec->lk_suppgid1;
1266         uc->uc_suppgids[1] = rec->lk_suppgid2;
1267
1268         attr->la_uid = rec->lk_fsuid;
1269         attr->la_gid = rec->lk_fsgid;
1270         rr->rr_fid1 = &rec->lk_fid1;
1271         rr->rr_fid2 = &rec->lk_fid2;
1272         attr->la_ctime = rec->lk_time;
1273         attr->la_mtime = rec->lk_time;
1274         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1275
1276         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1277         if (rc < 0)
1278                 RETURN(rc);
1279
1280         rc = mdt_dlmreq_unpack(info);
1281
1282         RETURN(rc);
1283 }
1284
1285 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1286 {
1287         struct lu_ucred         *uc  = mdt_ucred(info);
1288         struct mdt_rec_unlink   *rec;
1289         struct md_attr          *ma = &info->mti_attr;
1290         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1291         struct mdt_reint_record *rr = &info->mti_rr;
1292         struct req_capsule      *pill = info->mti_pill;
1293         int rc;
1294         ENTRY;
1295
1296         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1297         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1298         if (rec == NULL)
1299                 RETURN(-EFAULT);
1300
1301         /* This prior initialization is needed for old_init_ucred_reint() */
1302         uc->uc_fsuid = rec->ul_fsuid;
1303         uc->uc_fsgid = rec->ul_fsgid;
1304         uc->uc_cap   = rec->ul_cap;
1305         uc->uc_suppgids[0] = rec->ul_suppgid1;
1306         uc->uc_suppgids[1] = -1;
1307
1308         attr->la_uid = rec->ul_fsuid;
1309         attr->la_gid = rec->ul_fsgid;
1310         rr->rr_fid1 = &rec->ul_fid1;
1311         rr->rr_fid2 = &rec->ul_fid2;
1312         attr->la_ctime = rec->ul_time;
1313         attr->la_mtime = rec->ul_time;
1314         attr->la_mode  = rec->ul_mode;
1315         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1316
1317         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1318         if (rc < 0)
1319                 RETURN(rc);
1320
1321         if (rec->ul_bias & MDS_VTX_BYPASS)
1322                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1323         else
1324                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1325
1326         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1327
1328         rc = mdt_dlmreq_unpack(info);
1329         RETURN(rc);
1330 }
1331
1332 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1333 {
1334         info->mti_spec.sp_rm_entry = 1;
1335         return mdt_unlink_unpack(info);
1336 }
1337
1338 static int mdt_rename_unpack(struct mdt_thread_info *info)
1339 {
1340         struct lu_ucred         *uc = mdt_ucred(info);
1341         struct mdt_rec_rename   *rec;
1342         struct md_attr          *ma = &info->mti_attr;
1343         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1344         struct mdt_reint_record *rr = &info->mti_rr;
1345         struct req_capsule      *pill = info->mti_pill;
1346         int rc;
1347         ENTRY;
1348
1349         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1350         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1351         if (rec == NULL)
1352                 RETURN(-EFAULT);
1353
1354         /* This prior initialization is needed for old_init_ucred_reint() */
1355         uc->uc_fsuid = rec->rn_fsuid;
1356         uc->uc_fsgid = rec->rn_fsgid;
1357         uc->uc_cap   = rec->rn_cap;
1358         uc->uc_suppgids[0] = rec->rn_suppgid1;
1359         uc->uc_suppgids[1] = rec->rn_suppgid2;
1360
1361         attr->la_uid = rec->rn_fsuid;
1362         attr->la_gid = rec->rn_fsgid;
1363         rr->rr_fid1 = &rec->rn_fid1;
1364         rr->rr_fid2 = &rec->rn_fid2;
1365         attr->la_ctime = rec->rn_time;
1366         attr->la_mtime = rec->rn_time;
1367         /* rename_tgt contains the mode already */
1368         attr->la_mode = rec->rn_mode;
1369         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1370
1371         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1372         if (rc < 0)
1373                 RETURN(rc);
1374
1375         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1376         if (rc < 0)
1377                 RETURN(rc);
1378
1379         if (rec->rn_bias & MDS_VTX_BYPASS)
1380                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1381         else
1382                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1383
1384         if (rec->rn_bias & MDS_RENAME_MIGRATE) {
1385                 req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
1386                 rc = mdt_close_handle_unpack(info);
1387                 if (rc < 0)
1388                         RETURN(rc);
1389                 info->mti_spec.sp_migrate_close = 1;
1390         }
1391
1392         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1393
1394
1395         rc = mdt_dlmreq_unpack(info);
1396
1397         RETURN(rc);
1398 }
1399
1400 /*
1401  * please see comment above LOV_MAGIC_V1_DEF
1402  */
1403 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata)
1404 {
1405         struct lov_user_md_v1   *v1 = eadata;
1406
1407         LASSERT(v1);
1408
1409         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1410                 if ((v1->lmm_magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
1411                         v1->lmm_magic |= LOV_MAGIC_DEF;
1412                 else if ((v1->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
1413                          __swab32(LOV_MAGIC_MAGIC))
1414                         v1->lmm_magic |= __swab32(LOV_MAGIC_DEF);
1415         }
1416 }
1417
1418 static int mdt_open_unpack(struct mdt_thread_info *info)
1419 {
1420         struct lu_ucred         *uc = mdt_ucred(info);
1421         struct mdt_rec_create   *rec;
1422         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1423         struct req_capsule      *pill = info->mti_pill;
1424         struct mdt_reint_record *rr   = &info->mti_rr;
1425         struct ptlrpc_request   *req  = mdt_info_req(info);
1426         struct md_op_spec       *sp   = &info->mti_spec;
1427         int rc;
1428         ENTRY;
1429
1430         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1431         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1432         if (rec == NULL)
1433                 RETURN(-EFAULT);
1434
1435         /* This prior initialization is needed for old_init_ucred_reint() */
1436         uc->uc_fsuid = rec->cr_fsuid;
1437         uc->uc_fsgid = rec->cr_fsgid;
1438         uc->uc_cap   = rec->cr_cap;
1439         uc->uc_suppgids[0] = rec->cr_suppgid1;
1440         uc->uc_suppgids[1] = rec->cr_suppgid2;
1441         uc->uc_umask = rec->cr_umask;
1442
1443         rr->rr_fid1   = &rec->cr_fid1;
1444         rr->rr_fid2   = &rec->cr_fid2;
1445         rr->rr_handle = &rec->cr_old_handle;
1446         attr->la_mode = rec->cr_mode;
1447         attr->la_rdev  = rec->cr_rdev;
1448         attr->la_uid   = rec->cr_fsuid;
1449         attr->la_gid   = rec->cr_fsgid;
1450         attr->la_ctime = rec->cr_time;
1451         attr->la_mtime = rec->cr_time;
1452         attr->la_atime = rec->cr_time;
1453         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1454                          LA_CTIME | LA_MTIME | LA_ATIME;
1455         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1456         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1457         /* Do not trigger ASSERTION if client miss to set such flags. */
1458         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1459                 RETURN(-EPROTO);
1460
1461         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1462
1463         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1464
1465         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1466                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1467                                                         RCL_CLIENT);
1468                 if (rr->rr_eadatalen > 0) {
1469                         rr->rr_eadata = req_capsule_client_get(pill,
1470                                                                &RMF_EADATA);
1471                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1472                         sp->u.sp_ea.eadata = rr->rr_eadata;
1473                         sp->no_create = !!req_is_replay(req);
1474                         mdt_fix_lov_magic(info, rr->rr_eadata);
1475                 }
1476
1477                 /*
1478                  * Client default md_size may be 0 right after client start,
1479                  * until all osc are connected, set here just some reasonable
1480                  * value to prevent misbehavior.
1481                  */
1482                 if (rr->rr_eadatalen == 0 &&
1483                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1484                         rr->rr_eadatalen = MIN_MD_SIZE;
1485         }
1486
1487         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1488                                     &sp->sp_cr_file_secctx,
1489                                     &sp->sp_cr_file_secctx_size);
1490
1491         RETURN(rc);
1492 }
1493
1494 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1495 {
1496         struct mdt_reint_record *rr     = &info->mti_rr;
1497         struct lu_ucred         *uc     = mdt_ucred(info);
1498         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1499         struct req_capsule      *pill   = info->mti_pill;
1500         struct mdt_rec_setxattr *rec;
1501         int                      rc;
1502         ENTRY;
1503
1504
1505         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1506                          sizeof(struct mdt_rec_reint));
1507
1508         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1509         if (rec == NULL)
1510                 RETURN(-EFAULT);
1511
1512         /* This prior initialization is needed for old_init_ucred_reint() */
1513         uc->uc_fsuid  = rec->sx_fsuid;
1514         uc->uc_fsgid  = rec->sx_fsgid;
1515         uc->uc_cap    = rec->sx_cap;
1516         uc->uc_suppgids[0] = rec->sx_suppgid1;
1517         uc->uc_suppgids[1] = -1;
1518
1519         rr->rr_opcode = rec->sx_opcode;
1520         rr->rr_fid1   = &rec->sx_fid;
1521         attr->la_valid = rec->sx_valid;
1522         attr->la_ctime = rec->sx_time;
1523         attr->la_size = rec->sx_size;
1524         attr->la_flags = rec->sx_flags;
1525
1526         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1527         if (rc < 0)
1528                 RETURN(rc);
1529
1530         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1531                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1532                                                         RCL_CLIENT);
1533                 if (rr->rr_eadatalen > 0) {
1534                         rr->rr_eadata = req_capsule_client_get(pill,
1535                                                                &RMF_EADATA);
1536                         if (rr->rr_eadata == NULL)
1537                                 RETURN(-EFAULT);
1538                 } else {
1539                         rr->rr_eadata = NULL;
1540                 }
1541         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1542                 CDEBUG(D_INFO, "no xattr data supplied\n");
1543                 RETURN(-EFAULT);
1544         }
1545
1546         if (mdt_dlmreq_unpack(info) < 0)
1547                 RETURN(-EPROTO);
1548
1549         RETURN(0);
1550 }
1551
1552
1553 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1554
1555 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1556         [REINT_SETATTR]  = mdt_setattr_unpack,
1557         [REINT_CREATE]   = mdt_create_unpack,
1558         [REINT_LINK]     = mdt_link_unpack,
1559         [REINT_UNLINK]   = mdt_unlink_unpack,
1560         [REINT_RENAME]   = mdt_rename_unpack,
1561         [REINT_OPEN]     = mdt_open_unpack,
1562         [REINT_SETXATTR] = mdt_setxattr_unpack,
1563         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1564         [REINT_MIGRATE]  = mdt_rename_unpack,
1565 };
1566
1567 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1568 {
1569         int rc;
1570         ENTRY;
1571
1572         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1573         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1574                 info->mti_rr.rr_opcode = op;
1575                 rc = mdt_reint_unpackers[op](info);
1576         } else {
1577                 CERROR("Unexpected opcode %d\n", op);
1578                 rc = -EFAULT;
1579         }
1580         RETURN(rc);
1581 }