lustre/mds/mds_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (c) 2003 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #define DEBUG_SUBSYSTEM S_MDS
  23
  24 #include <linux/config.h>
  25 #include <linux/module.h>
  26 #include <linux/kernel.h>
  27 #include <linux/mm.h>
  28 #include <linux/string.h>
  29 #include <linux/stat.h>
  30 #include <linux/errno.h>
  31 #include <linux/version.h>
  32 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  33 # include <linux/locks.h>   // for wait_on_buffer
  34 #else
  35 # include <linux/buffer_head.h>   // for wait_on_buffer
  36 #endif
  37 #include <linux/unistd.h>
  38
  39 #include <asm/system.h>
  40 #include <asm/uaccess.h>
  41
  42 #include <linux/fs.h>
  43 #include <linux/stat.h>
  44 #include <asm/uaccess.h>
  45 #include <linux/slab.h>
  46 #include <asm/segment.h>
  47 #include <linux/random.h>
  48
  49 #include <linux/obd_support.h>
  50 #include <linux/lustre_lib.h>
  51 #include <linux/lustre_sec.h>
  52 #include <linux/lustre_ucache.h>
  53 #include <linux/lustre_gs.h>
  54 #include <linux/lustre_fsfilt.h>
  55
  56 #include "mds_internal.h"
  57
  58 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,4)
  59 struct group_info *groups_alloc(int ngroups)
  60 {
  61         struct group_info *ginfo;
  62
  63         LASSERT(ngroups <= NGROUPS_SMALL);
  64
  65         OBD_ALLOC(ginfo, sizeof(*ginfo) + 1 * sizeof(gid_t *));
  66         if (!ginfo)
  67                 return NULL;
  68         ginfo->ngroups = ngroups;
  69         ginfo->nblocks = 1;
  70         ginfo->blocks[0] = ginfo->small_block;
  71         atomic_set(&ginfo->usage, 1);
  72
  73         return ginfo;
  74 }
  75
  76 void groups_free(struct group_info *ginfo)
  77 {
  78         LASSERT(ginfo->ngroups <= NGROUPS_SMALL);
  79         LASSERT(ginfo->nblocks == 1);
  80         LASSERT(ginfo->blocks[0] == ginfo->small_block);
  81
  82         OBD_FREE(ginfo, sizeof(*ginfo) + 1 * sizeof(gid_t *));
  83 }
  84
  85 /* for 2.4 the group number is small, so simply search the
  86  * whole array.
  87  */
  88 int groups_search(struct group_info *ginfo, gid_t grp)
  89 {
  90         int i;
  91
  92         if (!ginfo)
  93                 return 0;
  94
  95         for (i = 0; i < ginfo->ngroups; i++)
  96                 if (GROUP_AT(ginfo, i) == grp)
  97                         return 1;
  98         return 0;
  99 }
 100
 101 #else /* >= 2.6.4 */
 102
 103 void groups_sort(struct group_info *ginfo)
 104 {
 105         int base, max, stride;
 106         int gidsetsize = ginfo->ngroups;
 107
 108         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
 109                 ; /* nothing */
 110         stride /= 3;
 111
 112         while (stride) {
 113                 max = gidsetsize - stride;
 114                 for (base = 0; base < max; base++) {
 115                         int left = base;
 116                         int right = left + stride;
 117                         gid_t tmp = GROUP_AT(ginfo, right);
 118
 119                         while (left >= 0 && GROUP_AT(ginfo, left) > tmp) {
 120                                 GROUP_AT(ginfo, right) =
 121                                     GROUP_AT(ginfo, left);
 122                                 right = left;
 123                                 left -= stride;
 124                         }
 125                         GROUP_AT(ginfo, right) = tmp;
 126                 }
 127                 stride /= 3;
 128         }
 129 }
 130
 131 int groups_search(struct group_info *ginfo, gid_t grp)
 132 {
 133         int left, right;
 134
 135         if (!ginfo)
 136                 return 0;
 137
 138         left = 0;
 139         right = ginfo->ngroups;
 140         while (left < right) {
 141                 int mid = (left + right) / 2;
 142                 int cmp = grp - GROUP_AT(ginfo, mid);
 143                 if (cmp > 0)
 144                         left = mid + 1;
 145                 else if (cmp < 0)
 146                         right = mid;
 147                 else
 148                         return 1;
 149         }
 150         return 0;
 151 }
 152 #endif
 153
 154 void groups_from_buffer(struct group_info *ginfo, __u32 *gids)
 155 {
 156         int i, ngroups = ginfo->ngroups;
 157
 158         for (i = 0; i < ginfo->nblocks; i++) {
 159                 int count = min(NGROUPS_PER_BLOCK, ngroups);
 160
 161                 memcpy(ginfo->blocks[i], gids, count * sizeof(__u32));
 162                 gids += NGROUPS_PER_BLOCK;
 163                 ngroups -= count;
 164         }
 165 }
 166
 167 void mds_pack_dentry2id(struct obd_device *obd,
 168                         struct lustre_id *id,
 169                         struct dentry *dentry,
 170                         int fid)
 171 {
 172         id_ino(id) = dentry->d_inum;
 173         id_gen(id) = dentry->d_generation;
 174
 175         if (fid) {
 176                 id_fid(id) = dentry->d_fid;
 177                 id_group(id) = dentry->d_mdsnum;
 178         }
 179 }
 180
 181 void mds_pack_dentry2body(struct obd_device *obd,
 182                           struct mds_body *b,
 183                           struct dentry *dentry,
 184                           int fid)
 185 {
 186         b->valid |= OBD_MD_FLID | OBD_MD_FLGENER |
 187                 OBD_MD_MDS;
 188
 189         if (fid)
 190                 b->valid |= OBD_MD_FID;
 191
 192         mds_pack_dentry2id(obd, &b->id1, dentry, fid);
 193 }
 194
 195 int mds_pack_inode2id(struct obd_device *obd,
 196                       struct lustre_id *id,
 197                       struct inode *inode,
 198                       int fid)
 199 {
 200         int rc = 0;
 201         ENTRY;
 202
 203         if (fid) {
 204                 /* we have to avoid deadlock. */
 205                 if (!down_trylock(&inode->i_sem)) {
 206                         rc = mds_read_inode_sid(obd, inode, id);
 207                         up(&inode->i_sem);
 208                 } else {
 209                         rc = mds_read_inode_sid(obd, inode, id);
 210                 }
 211         }
 212
 213         if (rc == 0) {
 214                 id_ino(id) = inode->i_ino;
 215                 id_gen(id) = inode->i_generation;
 216                 id_type(id) = (S_IFMT & inode->i_mode);
 217         }
 218         RETURN(rc);
 219 }
 220
 221 void mds_inode2id(struct obd_device *obd, struct lustre_id *id,
 222                   struct inode *inode, __u64 fid)
 223 {
 224         struct mds_obd *mds = &obd->u.mds;
 225         ENTRY;
 226
 227         LASSERT(inode != NULL);
 228         LASSERT(id != NULL);
 229         LASSERT(fid != 0);
 230
 231         id_fid(id) = fid;
 232         id_ino(id) = inode->i_ino;
 233         id_group(id) = mds->mds_num;
 234         id_gen(id) = inode->i_generation;
 235         id_type(id) = (S_IFMT & inode->i_mode);
 236
 237         EXIT;
 238 }
 239
 240 int mds_pack_gskey(struct obd_device *obd, struct lustre_msg *repmsg,
 241                   int *offset, struct mds_body *body, struct inode *inode)
 242 {
 243         struct crypto_key_md *md_key;
 244         struct crypto_key *ckey;
 245         __u32 buflen, *sizep;
 246         void *buf;
 247         int size, rc = 0;
 248         ENTRY;
 249
 250         sizep = lustre_msg_buf(repmsg, (*offset)++, 4);
 251         if (!sizep) {
 252                 CERROR("can't locate returned ckey size buf\n");
 253                 RETURN(-EPROTO);
 254         }
 255         *sizep = cpu_to_le32(sizeof(*ckey));
 256
 257         OBD_ALLOC(md_key, sizeof(*md_key));
 258         if (!md_key)
 259                 RETURN(-ENOMEM);
 260
 261         buflen = repmsg->buflens[*offset];
 262         buf = lustre_msg_buf(repmsg, (*offset)++, buflen);
 263
 264         size = fsfilt_get_md(obd, inode, md_key, sizeof(*md_key),
 265                              EA_KEY);
 266         if (size <= 0) {
 267                 if (size < 0)
 268                         CERROR("Can not get gskey from MDS ino %lu rc %d\n",
 269                                 inode->i_ino, size);
 270                 (*offset) += 2; /* XXX: ignore crypto in case size == 0 */
 271                 GOTO(out, rc = size);
 272         }
 273         if (le32_to_cpu(md_key->md_magic) != MD_KEY_MAGIC) {
 274                 CDEBUG(D_INFO, "given match %x != magic %x\n",
 275                        md_key->md_magic, MD_KEY_MAGIC);
 276                 GOTO(out, rc = 0);
 277         }
 278
 279         CDEBUG(D_INFO, "get key %s mac %s for ino %lu  size %d \n",
 280                md_key->md_ck.ck_key, md_key->md_ck.ck_mac, inode->i_ino, size);
 281         ckey=(struct crypto_key*)buf;
 282
 283         memcpy(ckey, &md_key->md_ck, sizeof(*ckey));
 284         body->valid |= OBD_MD_FLKEY;
 285 out:
 286         OBD_FREE(md_key, sizeof(*md_key));
 287         RETURN(rc);
 288 }
 289
 290 static int mds_get_gskey(struct inode *inode, struct crypto_key *ckey)
 291 {
 292         LASSERT(ckey);
 293         /*tmp create gs key here*/
 294         LASSERT(ckey->ck_type == MKS_TYPE);
 295         get_random_bytes(ckey->ck_key, KEY_SIZE);
 296         RETURN(0);
 297 }
 298
 299 int mds_set_gskey(struct obd_device *obd, void *handle,
 300                   struct inode *inode, void *key, int key_len,
 301                   int valid)
 302 {
 303         struct crypto_key_md *md_key = NULL;
 304         struct crypto_key *ckey = (struct crypto_key *)key;
 305         int rc = 0;
 306         ENTRY;
 307
 308         if (!ckey)
 309                 RETURN(0);
 310
 311         LASSERT(ckey->ck_type == MKS_TYPE || ckey->ck_type == GKS_TYPE);
 312
 313         OBD_ALLOC(md_key, sizeof(*md_key));
 314         if (ckey->ck_type == MKS_TYPE)
 315                 mds_get_gskey(inode, ckey);
 316
 317         rc = fsfilt_get_md(obd, inode, md_key, sizeof(*md_key),
 318                            EA_KEY);
 319         if (rc < 0)
 320                 GOTO(free, rc);
 321         LASSERT(le32_to_cpu(md_key->md_magic) == MD_KEY_MAGIC ||
 322                 md_key->md_magic == 0);
 323
 324         if (le32_to_cpu(md_key->md_magic) == MD_KEY_MAGIC) {
 325                 CDEBUG(D_INFO, "reset key %s mac %s", md_key->md_ck.ck_mac,
 326                        md_key->md_ck.ck_key);
 327         }
 328
 329         md_key->md_magic = cpu_to_le32(MD_KEY_MAGIC);
 330                 /*get key and mac from request buffer*/
 331         if (valid & ATTR_MAC) {
 332                 memcpy(md_key->md_ck.ck_mac, ckey->ck_mac, MAC_SIZE);
 333                         CDEBUG(D_INFO, "set mac %s for ino %lu \n",
 334                                         md_key->md_ck.ck_mac, inode->i_ino);
 335         }
 336         if (valid & ATTR_KEY) {
 337                 memcpy(md_key->md_ck.ck_key, ckey->ck_key, KEY_SIZE);
 338                        CDEBUG(D_INFO, "set key %s for ino %lu \n",
 339                                         md_key->md_ck.ck_key, inode->i_ino);
 340         }
 341         rc = fsfilt_set_md(obd, inode, handle, md_key, sizeof(*md_key), EA_KEY);
 342 free:
 343         if (md_key)
 344                 OBD_FREE(md_key, sizeof(*md_key));
 345         RETURN(rc);
 346 }
 347
 348 int mds_set_crypto_type(struct obd_device *obd, void *val, __u32 vallen)
 349 {
 350         struct mds_obd *mds = &obd->u.mds;
 351         ENTRY;
 352         if (vallen >= strlen("mks") &&
 353              memcmp(val, "mks", vallen) == 0) {
 354                 mds->mds_crypto_type = MKS_TYPE;
 355                 CDEBUG(D_IOCTL, "mks type\n");
 356         }
 357         if (vallen >= strlen("gks") &&
 358              memcmp(val, "gks", vallen) == 0) {
 359                 mds->mds_crypto_type = GKS_TYPE;
 360                 CDEBUG(D_IOCTL, "gks type \n");
 361         }
 362         RETURN(0);
 363 }
 364
 365 /* Note that we can copy all of the fields, just some will not be "valid" */
 366 void mds_pack_inode2body(struct obd_device *obd, struct mds_body *b,
 367                          struct inode *inode, int fid)
 368 {
 369         b->valid |= OBD_MD_FLID | OBD_MD_FLCTIME | OBD_MD_FLUID |
 370                 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLTYPE |
 371                 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLGENER |
 372                 OBD_MD_FLATIME | OBD_MD_FLMTIME; /* bug 2020 */
 373
 374         if (!S_ISREG(inode->i_mode)) {
 375                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
 376                         OBD_MD_FLATIME | OBD_MD_FLMTIME |
 377                         OBD_MD_FLRDEV;
 378         }
 379         b->atime = LTIME_S(inode->i_atime);
 380         b->mtime = LTIME_S(inode->i_mtime);
 381         b->ctime = LTIME_S(inode->i_ctime);
 382         b->mode = inode->i_mode;
 383         b->size = inode->i_size;
 384         b->blocks = inode->i_blocks;
 385         b->uid = inode->i_uid;
 386         b->gid = inode->i_gid;
 387         b->flags = inode->i_flags;
 388         b->rdev = inode->i_rdev;
 389
 390         /* Return the correct link count for orphan inodes */
 391         if (mds_inode_is_orphan(inode)) {
 392                 b->nlink = 0;
 393         } else if (S_ISDIR(inode->i_mode)) {
 394                 b->nlink = 1;
 395         } else {
 396                 b->nlink = inode->i_nlink;
 397         }
 398
 399         if (fid)
 400                 b->valid |= OBD_MD_FID;
 401
 402         mds_pack_inode2id(obd, &b->id1, inode, fid);
 403 }
 404
 405 /* unpacking */
 406 static int mds_setattr_unpack(struct ptlrpc_request *req, int offset,
 407                               struct mds_update_record *r)
 408 {
 409         struct iattr *attr = &r->ur_iattr;
 410         struct mds_rec_setattr *rec;
 411         ENTRY;
 412
 413         rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
 414                                  lustre_swab_mds_rec_setattr);
 415         if (rec == NULL)
 416                 RETURN (-EFAULT);
 417
 418         r->ur_id1 = &rec->sa_id;
 419         r->ur_flags = rec->sa_flags;
 420         attr->ia_valid = rec->sa_valid;
 421         attr->ia_mode = rec->sa_mode;
 422         attr->ia_uid = rec->sa_uid;
 423         attr->ia_gid = rec->sa_gid;
 424         attr->ia_size = rec->sa_size;
 425         LTIME_S(attr->ia_atime) = rec->sa_atime;
 426         LTIME_S(attr->ia_mtime) = rec->sa_mtime;
 427         LTIME_S(attr->ia_ctime) = rec->sa_ctime;
 428         attr->ia_attr_flags = rec->sa_attr_flags;
 429
 430         LASSERT_REQSWAB(req, offset + 1);
 431         if (req->rq_reqmsg->bufcount > offset + 1) {
 432                 r->ur_eadata = lustre_msg_buf(req->rq_reqmsg,
 433                                               offset + 1, 0);
 434                 if (r->ur_eadata == NULL)
 435                         RETURN (-EFAULT);
 436                 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 1];
 437         }
 438
 439         if (req->rq_reqmsg->bufcount > offset + 2) {
 440                 r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
 441                 if (r->ur_ea2data == NULL)
 442                         RETURN (-EFAULT);
 443
 444                 r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 2];
 445         }
 446
 447         if (req->rq_reqmsg->bufcount > offset + 3) {
 448                 r->ur_ea3data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
 449                 if (r->ur_ea3data == NULL)
 450                         RETURN (-EFAULT);
 451
 452                 r->ur_ea3datalen = req->rq_reqmsg->buflens[offset + 3];
 453         }
 454
 455         RETURN(0);
 456 }
 457
 458 static int mds_create_unpack(struct ptlrpc_request *req, int offset,
 459                              struct mds_update_record *r)
 460 {
 461         struct mds_rec_create *rec;
 462         ENTRY;
 463
 464         rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
 465                                  lustre_swab_mds_rec_create);
 466         if (rec == NULL)
 467                 RETURN (-EFAULT);
 468
 469         r->ur_id1 = &rec->cr_id;
 470         r->ur_id2 = &rec->cr_replayid;
 471         r->ur_mode = rec->cr_mode;
 472         r->ur_rdev = rec->cr_rdev;
 473         r->ur_time = rec->cr_time;
 474         r->ur_flags = rec->cr_flags;
 475
 476         LASSERT_REQSWAB(req, offset + 1);
 477         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 478         if (r->ur_name == NULL)
 479                 RETURN(-EFAULT);
 480         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 481
 482         LASSERT_REQSWAB(req, offset + 2);
 483         if (req->rq_reqmsg->bufcount > offset + 2) {
 484                 if (S_ISLNK(r->ur_mode)) {
 485                         r->ur_tgt = lustre_msg_string(req->rq_reqmsg,
 486                                                       offset + 2, 0);
 487                         if (r->ur_tgt == NULL)
 488                                 RETURN(-EFAULT);
 489                         r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
 490                 } else if (S_ISDIR(r->ur_mode) ) {
 491                         /* Stripe info for mkdir - just a 16bit integer */
 492                         if (req->rq_reqmsg->buflens[offset + 2] != 2) {
 493                                 CERROR("mkdir stripe info does not match "
 494                                        "expected size %d vs 2\n",
 495                                        req->rq_reqmsg->buflens[offset + 2]);
 496                                 RETURN(-EINVAL);
 497                         }
 498                         r->ur_eadata = lustre_swab_buf(req->rq_reqmsg,
 499                                                        offset + 2, 2,
 500                                                        __swab16s);
 501                         r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
 502                 } else if (S_ISREG(r->ur_mode)){
 503                         r->ur_eadata = lustre_msg_buf(req->rq_reqmsg,
 504                                                       offset + 2, 0);
 505                         r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
 506                 } else {
 507                         /* Hm, no other users so far? */
 508                         LBUG();
 509                 }
 510         }
 511         RETURN(0);
 512 }
 513
 514 static int mds_link_unpack(struct ptlrpc_request *req, int offset,
 515                            struct mds_update_record *r)
 516 {
 517         struct mds_rec_link *rec;
 518         ENTRY;
 519
 520         rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
 521                                  lustre_swab_mds_rec_link);
 522         if (rec == NULL)
 523                 RETURN(-EFAULT);
 524
 525         r->ur_id1 = &rec->lk_id1;
 526         r->ur_id2 = &rec->lk_id2;
 527         r->ur_time = rec->lk_time;
 528         r->ur_flags = rec->lk_flags;
 529
 530         LASSERT_REQSWAB(req, offset + 1);
 531         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 532         if (r->ur_name == NULL)
 533                 RETURN(-EFAULT);
 534         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 535         RETURN(0);
 536 }
 537
 538 static int mds_unlink_unpack(struct ptlrpc_request *req, int offset,
 539                              struct mds_update_record *r)
 540 {
 541         struct mds_rec_unlink *rec;
 542         ENTRY;
 543
 544         rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
 545                                  lustre_swab_mds_rec_unlink);
 546         if (rec == NULL)
 547                 RETURN(-EFAULT);
 548
 549         r->ur_mode = rec->ul_mode;
 550         r->ur_id1 = &rec->ul_id1;
 551         r->ur_id2 = &rec->ul_id2;
 552         r->ur_time = rec->ul_time;
 553         r->ur_flags = rec->ul_flags;
 554
 555         LASSERT_REQSWAB(req, offset + 1);
 556         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 557         if (r->ur_name == NULL)
 558                 RETURN(-EFAULT);
 559         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 560         RETURN(0);
 561 }
 562
 563 static int mds_rename_unpack(struct ptlrpc_request *req, int offset,
 564                              struct mds_update_record *r)
 565 {
 566         struct mds_rec_rename *rec;
 567         ENTRY;
 568
 569         rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
 570                                  lustre_swab_mds_rec_rename);
 571         if (rec == NULL)
 572                 RETURN(-EFAULT);
 573
 574         r->ur_id1 = &rec->rn_id1;
 575         r->ur_id2 = &rec->rn_id2;
 576         r->ur_time = rec->rn_time;
 577         r->ur_flags = rec->rn_flags;
 578
 579         LASSERT_REQSWAB(req, offset + 1);
 580         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 581         if (r->ur_name == NULL)
 582                 RETURN(-EFAULT);
 583         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 584
 585         LASSERT_REQSWAB(req, offset + 2);
 586         r->ur_tgt = lustre_msg_string(req->rq_reqmsg, offset + 2, 0);
 587         if (r->ur_tgt == NULL)
 588                 RETURN(-EFAULT);
 589         r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
 590         RETURN(0);
 591 }
 592
 593 static int mds_open_unpack(struct ptlrpc_request *req, int offset,
 594                            struct mds_update_record *r)
 595 {
 596         struct mds_rec_create *rec;
 597         ENTRY;
 598
 599         rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
 600                                  lustre_swab_mds_rec_create);
 601         if (rec == NULL)
 602                 RETURN(-EFAULT);
 603
 604         r->ur_id1 = &rec->cr_id;
 605         r->ur_id2 = &rec->cr_replayid;
 606         r->ur_mode = rec->cr_mode;
 607         r->ur_rdev = rec->cr_rdev;
 608         r->ur_time = rec->cr_time;
 609         r->ur_flags = rec->cr_flags;
 610
 611         LASSERT_REQSWAB(req, offset + 1);
 612         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 613
 614         if (r->ur_name == NULL)
 615                 RETURN(-EFAULT);
 616         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 617
 618         LASSERT_REQSWAB(req, offset + 2);
 619         if (req->rq_reqmsg->bufcount > offset + 2) {
 620                 r->ur_eadata = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
 621                 if (r->ur_eadata == NULL)
 622                         RETURN(-EFAULT);
 623                 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
 624         }
 625
 626         if (rec->cr_flags & MDS_OPEN_HAS_KEY) {
 627                 LASSERT(req->rq_reqmsg->bufcount > offset + 3);
 628                 r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
 629                 r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 3];
 630         }
 631         RETURN(0);
 632 }
 633
 634 typedef int (*update_unpacker)(struct ptlrpc_request *req, int offset,
 635                                struct mds_update_record *r);
 636
 637 static update_unpacker mds_unpackers[REINT_MAX + 1] = {
 638         [REINT_SETATTR] mds_setattr_unpack,
 639         [REINT_CREATE] mds_create_unpack,
 640         [REINT_LINK] mds_link_unpack,
 641         [REINT_UNLINK] mds_unlink_unpack,
 642         [REINT_RENAME] mds_rename_unpack,
 643         [REINT_OPEN] mds_open_unpack,
 644 };
 645
 646 int mds_update_unpack(struct ptlrpc_request *req, int offset,
 647                       struct mds_update_record *rec)
 648 {
 649         __u32 *opcodep;
 650         __u32  opcode;
 651         int rc;
 652         ENTRY;
 653
 654         /* NB don't lustre_swab_reqbuf() here. We're just taking a peek and we
 655          * want to leave it to the specific unpacker once we've identified the
 656          * message type. */
 657         opcodep = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*opcodep));
 658         if (opcodep == NULL)
 659                 RETURN(-EFAULT);
 660
 661         opcode = *opcodep;
 662         if (lustre_msg_swabbed(req->rq_reqmsg))
 663                 __swab32s(&opcode);
 664
 665         if (opcode > REINT_MAX ||
 666             mds_unpackers[opcode] == NULL) {
 667                 CERROR("Unexpected opcode %d\n", opcode);
 668                 RETURN(-EFAULT);
 669         }
 670
 671         rec->ur_id1 = NULL;
 672         rec->ur_id2 = NULL;
 673         rec->ur_opcode = opcode;
 674
 675         rc = mds_unpackers[opcode](req, offset, rec);
 676
 677 #if CRAY_PORTALS
 678         rec->ur_fsuid = req->rq_uid;
 679 #endif
 680         RETURN(rc);
 681 }
 682
 683 /*
 684  * here we take simple rule: once uid/fsuid is root, we also squash
 685  * the gid/fsgid, don't care setuid/setgid attributes.
 686  */
 687 static
 688 int mds_squash_root(struct mds_obd *mds, struct mds_req_sec_desc *rsd,
 689                     ptl_nid_t *peernid)
 690 {
 691         if (!mds->mds_squash_uid || *peernid == mds->mds_nosquash_nid)
 692                 return 0;
 693
 694         if (rsd->rsd_uid && rsd->rsd_fsuid)
 695                 return 0;
 696
 697         CDEBUG(D_SEC, "squash req from "LPX64":"
 698                "(%u:%u-%u:%u/%x)=>(%u:%u-%u:%u/%x)\n", *peernid,
 699                 rsd->rsd_uid, rsd->rsd_gid,
 700                 rsd->rsd_fsuid, rsd->rsd_fsgid, rsd->rsd_cap,
 701                 rsd->rsd_uid ? rsd->rsd_uid : mds->mds_squash_uid,
 702                 rsd->rsd_uid ? rsd->rsd_gid : mds->mds_squash_gid,
 703                 rsd->rsd_fsuid ? rsd->rsd_fsuid : mds->mds_squash_uid,
 704                 rsd->rsd_fsuid ? rsd->rsd_fsgid : mds->mds_squash_gid,
 705                 rsd->rsd_cap & ~CAP_FS_MASK);
 706
 707         if (rsd->rsd_uid == 0) {
 708                 rsd->rsd_uid = mds->mds_squash_uid;
 709                 rsd->rsd_gid = mds->mds_squash_gid;
 710         }
 711         if (rsd->rsd_fsuid == 0) {
 712                 rsd->rsd_fsuid = mds->mds_squash_uid;
 713                 rsd->rsd_fsgid = mds->mds_squash_gid;
 714         }
 715         rsd->rsd_cap &= ~CAP_FS_MASK;
 716
 717         return 1;
 718 }
 719
 720 /********************************
 721  * MDS uid/gid mapping handling *
 722  ********************************/
 723
 724 static
 725 struct mds_idmap_entry* idmap_alloc_entry(__u32 rmt_id, __u32 lcl_id)
 726 {
 727         struct mds_idmap_entry *e;
 728
 729         OBD_ALLOC(e, sizeof(*e));
 730         if (!e)
 731                 return NULL;
 732
 733         INIT_LIST_HEAD(&e->rmt_hash);
 734         INIT_LIST_HEAD(&e->lcl_hash);
 735         atomic_set(&e->refcount, 1);
 736         e->rmt_id = rmt_id;
 737         e->lcl_id = lcl_id;
 738
 739         return e;
 740 }
 741
 742 void idmap_free_entry(struct mds_idmap_entry *e)
 743 {
 744         if (!list_empty(&e->rmt_hash))
 745                 list_del(&e->rmt_hash);
 746         if (!list_empty(&e->lcl_hash))
 747                 list_del(&e->lcl_hash);
 748         OBD_FREE(e, sizeof(*e));
 749 }
 750
 751 static
 752 int idmap_insert_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
 753                        struct mds_idmap_entry *new, const char *warn_msg)
 754 {
 755         struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(new->rmt_id)];
 756         struct list_head *lcl_head = &lcl_hash[MDS_IDMAP_HASHFUNC(new->lcl_id)];
 757         struct mds_idmap_entry *e;
 758
 759         list_for_each_entry(e, rmt_head, rmt_hash) {
 760                 if (e->rmt_id == new->rmt_id &&
 761                     e->lcl_id == new->lcl_id) {
 762                         atomic_inc(&e->refcount);
 763                         return 1;
 764                 }
 765                 if (e->rmt_id == new->rmt_id && warn_msg)
 766                         CWARN("%s: rmt id %u already map to %u (new %u)\n",
 767                               warn_msg, e->rmt_id, e->lcl_id, new->lcl_id);
 768                 if (e->lcl_id == new->lcl_id && warn_msg)
 769                         CWARN("%s: lcl id %u already be mapped from %u "
 770                               "(new %u)\n", warn_msg,
 771                               e->lcl_id, e->rmt_id, new->rmt_id);
 772         }
 773
 774         list_add_tail(rmt_head, &new->rmt_hash);
 775         list_add_tail(lcl_head, &new->lcl_hash);
 776         return 0;
 777 }
 778
 779 static
 780 int idmap_remove_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
 781                        __u32 rmt_id, __u32 lcl_id)
 782 {
 783         struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(rmt_id)];
 784         struct mds_idmap_entry *e;
 785
 786         list_for_each_entry(e, rmt_head, rmt_hash) {
 787                 if (e->rmt_id == rmt_id && e->lcl_id == lcl_id) {
 788                         if (atomic_dec_and_test(&e->refcount)) {
 789                                 list_del(&e->rmt_hash);
 790                                 list_del(&e->lcl_hash);
 791                                 OBD_FREE(e, sizeof(*e));
 792                                 return 0;
 793                         } else
 794                                 return 1;
 795                 }
 796         }
 797         return -ENOENT;
 798 }
 799
 800 int mds_idmap_add(struct mds_idmap_table *tbl,
 801                   uid_t rmt_uid, uid_t lcl_uid,
 802                   gid_t rmt_gid, gid_t lcl_gid)
 803 {
 804         struct mds_idmap_entry *ue, *ge;
 805         ENTRY;
 806
 807         if (!tbl)
 808                 RETURN(-EPERM);
 809
 810         ue = idmap_alloc_entry(rmt_uid, lcl_uid);
 811         if (!ue)
 812                 RETURN(-ENOMEM);
 813         ge = idmap_alloc_entry(rmt_gid, lcl_gid);
 814         if (!ge) {
 815                 idmap_free_entry(ue);
 816                 RETURN(-ENOMEM);
 817         }
 818
 819         spin_lock(&tbl->mit_lock);
 820
 821         if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
 822                                tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
 823                                ue, "UID mapping")) {
 824                 idmap_free_entry(ue);
 825         }
 826
 827         if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
 828                                tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
 829                                ge, "GID mapping")) {
 830                 idmap_free_entry(ge);
 831         }
 832
 833         spin_unlock(&tbl->mit_lock);
 834         RETURN(0);
 835 }
 836
 837 int mds_idmap_del(struct mds_idmap_table *tbl,
 838                   uid_t rmt_uid, uid_t lcl_uid,
 839                   gid_t rmt_gid, gid_t lcl_gid)
 840 {
 841         ENTRY;
 842
 843         if (!tbl)
 844                 RETURN(0);
 845
 846         spin_lock(&tbl->mit_lock);
 847         idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
 848                            tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
 849                            rmt_uid, lcl_uid);
 850         idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
 851                            tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
 852                            rmt_gid, lcl_gid);
 853         spin_unlock(&tbl->mit_lock);
 854         RETURN(0);
 855 }
 856
 857 static
 858 __u32 idmap_lookup_id(struct list_head *hash, int reverse, __u32 id)
 859 {
 860         struct list_head *head = &hash[MDS_IDMAP_HASHFUNC(id)];
 861         struct mds_idmap_entry *e;
 862
 863         if (!reverse) {
 864                 list_for_each_entry(e, head, rmt_hash) {
 865                         if (e->rmt_id == id)
 866                                 return e->lcl_id;
 867                 }
 868                 return MDS_IDMAP_NOTFOUND;
 869         } else {
 870                 list_for_each_entry(e, head, lcl_hash) {
 871                         if (e->lcl_id == id)
 872                                 return e->rmt_id;
 873                 }
 874                 return MDS_IDMAP_NOTFOUND;
 875         }
 876 }
 877
 878 int mds_idmap_lookup_uid(struct mds_idmap_table *tbl, int reverse, uid_t uid)
 879 {
 880         struct list_head *hash;
 881
 882         if (!tbl)
 883                 return MDS_IDMAP_NOTFOUND;
 884
 885         if (!reverse)
 886                 hash = tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX];
 887         else
 888                 hash = tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX];
 889
 890         spin_lock(&tbl->mit_lock);
 891         uid = idmap_lookup_id(hash, reverse, uid);
 892         spin_unlock(&tbl->mit_lock);
 893
 894         return uid;
 895 }
 896
 897 int mds_idmap_lookup_gid(struct mds_idmap_table *tbl, int reverse, gid_t gid)
 898 {
 899         struct list_head *hash;
 900
 901         if (!tbl)
 902                 return MDS_IDMAP_NOTFOUND;
 903
 904         if (!reverse)
 905                 hash = tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX];
 906         else
 907                 hash = tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX];
 908
 909         spin_lock(&tbl->mit_lock);
 910         gid = idmap_lookup_id(hash, reverse, gid);
 911         spin_unlock(&tbl->mit_lock);
 912
 913         return gid;
 914 }
 915
 916 struct mds_idmap_table *mds_idmap_alloc()
 917 {
 918         struct mds_idmap_table *tbl;
 919         int i, j;
 920
 921         OBD_ALLOC(tbl, sizeof(*tbl));
 922         if (!tbl)
 923                 return NULL;
 924
 925         spin_lock_init(&tbl->mit_lock);
 926         for (i = 0; i < MDS_IDMAP_N_HASHES; i++)
 927                 for (j = 0; j < MDS_IDMAP_HASHSIZE; j++)
 928                         INIT_LIST_HEAD(&tbl->mit_idmaps[i][j]);
 929
 930         return tbl;
 931 }
 932
 933 static void idmap_clear_rmt_hash(struct list_head *list)
 934 {
 935         struct mds_idmap_entry *e;
 936         int i;
 937
 938         for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
 939                 while (!list_empty(&list[i])) {
 940                         e = list_entry(list[i].next, struct mds_idmap_entry,
 941                                        rmt_hash);
 942                         idmap_free_entry(e);
 943                 }
 944         }
 945 }
 946
 947 void mds_idmap_free(struct mds_idmap_table *tbl)
 948 {
 949         int i;
 950
 951         spin_lock(&tbl->mit_lock);
 952         idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX]);
 953         idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX]);
 954
 955         /* paranoid checking */
 956         for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
 957                 LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX][i]));
 958                 LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX][i]));
 959         }
 960         spin_unlock(&tbl->mit_lock);
 961
 962         OBD_FREE(tbl, sizeof(*tbl));
 963 }
 964
 965 /*********************************
 966  * helpers doing mapping for MDS *
 967  *********************************/
 968
 969 /*
 970  * we allow remote setuid/setgid to an "authencated" one,
 971  * this policy probably change later.
 972  */
 973 static
 974 int mds_req_secdesc_do_map(struct mds_export_data *med,
 975                            struct mds_req_sec_desc *rsd)
 976 {
 977         struct mds_idmap_table *idmap = med->med_idmap;
 978         uid_t uid, fsuid;
 979         gid_t gid, fsgid;
 980
 981         uid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_uid);
 982         if (uid == MDS_IDMAP_NOTFOUND) {
 983                 CERROR("can't find map for uid %u\n", rsd->rsd_uid);
 984                 return -EPERM;
 985         }
 986
 987         if (rsd->rsd_uid == rsd->rsd_fsuid)
 988                 fsuid = uid;
 989         else {
 990                 fsuid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_fsuid);
 991                 if (fsuid == MDS_IDMAP_NOTFOUND) {
 992                         CERROR("can't find map for fsuid %u\n", rsd->rsd_fsuid);
 993                         return -EPERM;
 994                 }
 995         }
 996
 997         gid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_gid);
 998         if (gid == MDS_IDMAP_NOTFOUND) {
 999                 CERROR("can't find map for gid %u\n", rsd->rsd_gid);
1000                 return -EPERM;
1001         }
1002
1003         if (rsd->rsd_gid == rsd->rsd_fsgid)
1004                 fsgid = gid;
1005         else {
1006                 fsgid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_fsgid);
1007                 if (fsgid == MDS_IDMAP_NOTFOUND) {
1008                         CERROR("can't find map for fsgid %u\n", rsd->rsd_fsgid);
1009                         return -EPERM;
1010                 }
1011         }
1012
1013         rsd->rsd_uid = uid;
1014         rsd->rsd_gid = gid;
1015         rsd->rsd_fsuid = fsuid;
1016         rsd->rsd_fsgid = fsgid;
1017
1018         return 0;
1019 }
1020
1021 void mds_body_do_reverse_map(struct mds_export_data *med,
1022                              struct mds_body *body)
1023 {
1024         uid_t uid;
1025         gid_t gid;
1026
1027         if (!med->med_remote)
1028                 return;
1029
1030         ENTRY;
1031         if (body->valid & OBD_MD_FLUID) {
1032                 uid = mds_idmap_lookup_uid(med->med_idmap, 1, body->uid);
1033                 if (uid == MDS_IDMAP_NOTFOUND) {
1034                         uid = med->med_nllu;
1035                         if (body->valid & OBD_MD_FLMODE) {
1036                                 body->mode = (body->mode & ~S_IRWXU) |
1037                                              ((body->mode & S_IRWXO) << 6);
1038                         }
1039                 }
1040                 body->uid = uid;
1041         }
1042         if (body->valid & OBD_MD_FLGID) {
1043                 gid = mds_idmap_lookup_gid(med->med_idmap, 1, body->gid);
1044                 if (gid == MDS_IDMAP_NOTFOUND) {
1045                         gid = med->med_nllg;
1046                         if (body->valid & OBD_MD_FLMODE) {
1047                                 body->mode = (body->mode & ~S_IRWXG) |
1048                                              ((body->mode & S_IRWXO) << 3);
1049                         }
1050                 }
1051                 body->gid = gid;
1052         }
1053
1054         EXIT;
1055 }
1056
1057 /*
1058  * return error if can't find mapping, it's a error so should not
1059  * fall into nllu/nllg.
1060  */
1061 int mds_remote_perm_do_reverse_map(struct mds_export_data *med,
1062                                    struct mds_remote_perm *perm)
1063 {
1064         uid_t uid;
1065         gid_t gid;
1066
1067         LASSERT(med->med_remote);
1068
1069         uid = mds_idmap_lookup_uid(med->med_idmap, 1, perm->mrp_auth_uid);
1070         if (uid == MDS_IDMAP_NOTFOUND) {
1071                 CERROR("no map for uid %u\n", perm->mrp_auth_uid);
1072                 return -EPERM;
1073         }
1074         gid = mds_idmap_lookup_gid(med->med_idmap, 1, perm->mrp_auth_gid);
1075         if (gid == MDS_IDMAP_NOTFOUND) {
1076                 CERROR("no map for uid %u\n", perm->mrp_auth_uid);
1077                 return -EPERM;
1078         }
1079
1080         perm->mrp_auth_uid = uid;
1081         perm->mrp_auth_gid = gid;
1082         return 0;
1083 }
1084
1085 /**********************
1086  * MDS ucred handling *
1087  **********************/
1088
1089 static inline void drop_ucred_ginfo(struct lvfs_ucred *ucred)
1090 {
1091         if (ucred->luc_ginfo) {
1092                 put_group_info(ucred->luc_ginfo);
1093                 ucred->luc_ginfo = NULL;
1094         }
1095 }
1096
1097 static inline void drop_ucred_lsd(struct lvfs_ucred *ucred)
1098 {
1099         if (ucred->luc_lsd) {
1100                 mds_put_lsd(ucred->luc_lsd);
1101                 ucred->luc_lsd = NULL;
1102         }
1103 }
1104
1105 /*
1106  * the heart of the uid/gid handling and security checking.
1107  *
1108  * root could set any group_info if we allowed setgroups, while
1109  * normal user only could 'reduce' their group members -- which
1110  * is somewhat expensive.
1111  *
1112  * authenticated as mds user (using mds service credential) could
1113  * bypass all checkings.
1114  */
1115 int mds_init_ucred(struct lvfs_ucred *ucred,
1116                    struct ptlrpc_request *req,
1117                    struct mds_req_sec_desc *rsd)
1118 {
1119         struct mds_obd *mds = &req->rq_export->exp_obd->u.mds;
1120         struct mds_export_data *med = &req->rq_export->u.eu_mds_data;
1121         struct lustre_sec_desc *lsd;
1122         ptl_nid_t peernid = req->rq_peer.peer_id.nid;
1123         struct group_info *gnew;
1124         unsigned int setuid, setgid, strong_sec, root_squashed;
1125         __u32 lsd_perms;
1126         ENTRY;
1127
1128         LASSERT(ucred);
1129         LASSERT(rsd);
1130         LASSERT(rsd->rsd_ngroups <= LUSTRE_MAX_GROUPS);
1131
1132         if (SEC_FLAVOR_MAJOR(req->rq_req_secflvr) == PTLRPCS_FLVR_MAJOR_GSS &&
1133             (SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_AUTH ||
1134              SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_PRIV))
1135                 strong_sec = 1;
1136         else
1137                 strong_sec = 0;
1138
1139         LASSERT(!(req->rq_remote_realm && !strong_sec));
1140
1141         if (strong_sec && req->rq_auth_uid == -1) {
1142                 CWARN("user not authenticated, deny access\n");
1143                 RETURN(-EPERM);
1144         }
1145
1146         /* sanity check: if we use strong authentication, we expect the
1147          * uid which client claimed is true.
1148          * not apply to special mds user .
1149          */
1150         if (!req->rq_auth_usr_mds && strong_sec) {
1151                 if (!med->med_remote) {
1152                         if (req->rq_auth_uid != rsd->rsd_uid) {
1153                                 CERROR("local client "LPU64": auth uid %u "
1154                                        "while client claim %u:%u/%u:%u\n",
1155                                        peernid, req->rq_auth_uid,
1156                                        rsd->rsd_uid, rsd->rsd_gid,
1157                                        rsd->rsd_fsuid, rsd->rsd_fsgid);
1158                                 RETURN(-EPERM);
1159                         }
1160                 } else {
1161                         if (req->rq_mapped_uid == MDS_IDMAP_NOTFOUND) {
1162                                 CWARN("no mapping found, deny\n");
1163                                 RETURN(-EPERM);
1164                         }
1165
1166                         if (mds_req_secdesc_do_map(med, rsd))
1167                                 RETURN(-EPERM);
1168
1169                         if (req->rq_mapped_uid != rsd->rsd_uid) {
1170                                 CERROR("remote client "LPU64": auth uid %u "
1171                                        "while client claim %u:%u/%u:%u\n",
1172                                        peernid, req->rq_auth_uid,
1173                                        rsd->rsd_uid, rsd->rsd_gid,
1174                                        rsd->rsd_fsuid, rsd->rsd_fsgid);
1175                                 RETURN(-EPERM);
1176                         }
1177                 }
1178         }
1179
1180         /* now LSD come into play */
1181         ucred->luc_ginfo = NULL;
1182         ucred->luc_lsd = lsd = mds_get_lsd(rsd->rsd_uid);
1183
1184         if (!lsd) {
1185                 CERROR("Deny access without LSD: uid %d\n", rsd->rsd_uid);
1186                 RETURN(-EPERM);
1187         }
1188
1189         lsd_perms = mds_lsd_get_perms(lsd, med->med_remote, 0, peernid);
1190
1191         /* check setuid/setgid permissions.
1192          * again not apply to special mds user.
1193          */
1194         if (!req->rq_auth_usr_mds) {
1195                 /* find out the setuid/setgid attempt */
1196                 setuid = (rsd->rsd_uid != rsd->rsd_fsuid);
1197                 setgid = (rsd->rsd_gid != rsd->rsd_fsgid ||
1198                           rsd->rsd_gid != lsd->lsd_gid);
1199
1200                 /* check permission of setuid */
1201                 if (setuid && !(lsd_perms & LSD_PERM_SETUID)) {
1202                         CWARN("mds blocked setuid attempt (%u -> %u) "
1203                               "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_fsuid,
1204                               peernid);
1205                         RETURN(-EPERM);
1206                 }
1207
1208                 /* check permission of setgid */
1209                 if (setgid && !(lsd_perms & LSD_PERM_SETGID)) {
1210                         CWARN("mds blocked setgid attempt (%u:%u/%u:%u -> %u) "
1211                               "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_gid,
1212                               rsd->rsd_fsuid, rsd->rsd_fsgid, lsd->lsd_gid,
1213                               peernid);
1214                         RETURN(-EPERM);
1215                 }
1216         }
1217
1218         root_squashed = mds_squash_root(mds, rsd, &peernid);
1219
1220         /* remove privilege for non-root user */
1221         if (rsd->rsd_fsuid)
1222                 rsd->rsd_cap &= ~CAP_FS_MASK;
1223
1224         /* by now every fields other than groups in rsd have been granted */
1225         ucred->luc_nid = peernid;
1226         ucred->luc_uid = rsd->rsd_uid;
1227         ucred->luc_gid = rsd->rsd_gid;
1228         ucred->luc_fsuid = rsd->rsd_fsuid;
1229         ucred->luc_fsgid = rsd->rsd_fsgid;
1230         ucred->luc_cap = rsd->rsd_cap;
1231
1232         /* don't use any supplementary group if we squashed root.
1233          * XXX The exact behavior of root_squash is not defined, we just
1234          * keep the reminder here */
1235         if (root_squashed)
1236                 RETURN(0);
1237
1238         /* install groups from LSD */
1239         if (lsd->lsd_ginfo) {
1240                 ucred->luc_ginfo = lsd->lsd_ginfo;
1241                 get_group_info(ucred->luc_ginfo);
1242         }
1243
1244         /* everything is done if we don't allow setgroups, or it is
1245          * from remote client (which implies forced to be no-setgroups).
1246          *
1247          * Note: remote user's supplementary groups sent along the request
1248          * (if any) are all ignored, but we make the mapped local user's
1249          * supplementary groups take effect.
1250          */
1251         if (med->med_remote || !(lsd_perms & LSD_PERM_SETGRP))
1252                 RETURN(0);
1253
1254         /* root could set any groups as he want (if allowed), normal
1255          * users only could reduce his group array.
1256          */
1257         if (ucred->luc_uid == 0) {
1258                 drop_ucred_ginfo(ucred);
1259
1260                 if (rsd->rsd_ngroups == 0)
1261                         RETURN(0);
1262
1263                 gnew = groups_alloc(rsd->rsd_ngroups);
1264                 if (!gnew) {
1265                         CERROR("out of memory\n");
1266                         drop_ucred_lsd(ucred);
1267                         RETURN(-ENOMEM);
1268                 }
1269                 groups_from_buffer(gnew, rsd->rsd_groups);
1270                 groups_sort(gnew); /* don't rely on client doing this */
1271
1272                 ucred->luc_ginfo = gnew;
1273         } else {
1274                 __u32 set = 0, cur = 0;
1275                 struct group_info *ginfo = ucred->luc_ginfo;
1276
1277                 if (!ginfo)
1278                         RETURN(0);
1279
1280                 /* Note: freeing a group_info count on 'nblocks' instead of
1281                  * 'ngroups', thus we can safely alloc enough buffer and reduce
1282                  * and ngroups number later.
1283                  */
1284                 gnew = groups_alloc(rsd->rsd_ngroups);
1285                 if (!gnew) {
1286                         CERROR("out of memory\n");
1287                         drop_ucred_ginfo(ucred);
1288                         drop_ucred_lsd(ucred);
1289                         RETURN(-ENOMEM);
1290                 }
1291
1292                 while (cur < rsd->rsd_ngroups) {
1293                         if (groups_search(ginfo, rsd->rsd_groups[cur])) {
1294                                 GROUP_AT(gnew, set) = rsd->rsd_groups[cur];
1295                                 set++;
1296                         }
1297                         cur++;
1298                 }
1299                 gnew->ngroups = set;
1300
1301                 put_group_info(ucred->luc_ginfo);
1302                 ucred->luc_ginfo = gnew;
1303         }
1304         RETURN(0);
1305 }
1306
1307 void mds_exit_ucred(struct lvfs_ucred *ucred)
1308 {
1309         ENTRY;
1310         drop_ucred_ginfo(ucred);
1311         drop_ucred_lsd(ucred);
1312         EXIT;
1313 }
1314