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
 259         buflen = repmsg->buflens[*offset];
 260         buf = lustre_msg_buf(repmsg, (*offset)++, buflen);
 261
 262         size = fsfilt_get_md(obd, inode, md_key, sizeof(*md_key),
 263                            EA_KEY);
 264         if (size <= 0) {
 265                 if (size < 0)
 266                         CERROR("Can not get gskey from MDS ino %lu rc %d\n",
 267                                 inode->i_ino, size);
 268                 (*offset) += 2; /* XXX: ignore crypto in case size == 0 */
 269                 GOTO(out, rc = size);
 270         }
 271         if (le32_to_cpu(md_key->md_magic) != MD_KEY_MAGIC) {
 272                 CDEBUG(D_INFO, "given match %x != magic %x\n",
 273                        md_key->md_magic, MD_KEY_MAGIC);
 274                 GOTO(out, rc = 0);
 275         }
 276
 277         CDEBUG(D_INFO, "get key %s mac %s for ino %lu  size %d \n",
 278                md_key->md_ck.ck_key, md_key->md_ck.ck_mac, inode->i_ino, size);
 279         ckey=(struct crypto_key*)buf;
 280
 281         memcpy(ckey, &md_key->md_ck, sizeof(*ckey));
 282         body->valid |= OBD_MD_FLKEY;
 283 out:
 284         OBD_FREE(md_key, sizeof(*md_key));
 285         RETURN(rc);
 286 }
 287
 288 static int mds_get_gskey(struct inode *inode, struct crypto_key *ckey)
 289 {
 290         LASSERT(ckey);
 291         /*tmp create gs key here*/
 292         LASSERT(ckey->ck_type == MKS_TYPE);
 293         get_random_bytes(ckey->ck_key, KEY_SIZE);
 294         RETURN(0);
 295 }
 296
 297 int mds_set_gskey(struct obd_device *obd, void *handle,
 298                   struct inode *inode, void *key, int key_len,
 299                   int valid)
 300 {
 301         struct crypto_key_md *md_key = NULL;
 302         struct crypto_key *ckey = (struct crypto_key *)key;
 303         int rc = 0;
 304         ENTRY;
 305
 306         if (!ckey)
 307                 RETURN(0);
 308
 309         LASSERT(ckey->ck_type == MKS_TYPE || ckey->ck_type == GKS_TYPE);
 310
 311         OBD_ALLOC(md_key, sizeof(*md_key));
 312         if (ckey->ck_type == MKS_TYPE) {
 313                 mds_get_gskey(inode, ckey);
 314         }
 315
 316         rc = fsfilt_get_md(obd, inode, md_key, sizeof(*md_key),
 317                            EA_KEY);
 318         if (rc < 0)
 319                 GOTO(free, rc);
 320         LASSERT(le32_to_cpu(md_key->md_magic) == MD_KEY_MAGIC ||
 321                 md_key->md_magic == 0);
 322
 323         if (le32_to_cpu(md_key->md_magic) == MD_KEY_MAGIC) {
 324                 CDEBUG(D_INFO, "reset key %s mac %s", md_key->md_ck.ck_mac,
 325                        md_key->md_ck.ck_key);
 326         }
 327
 328         md_key->md_magic = cpu_to_le32(MD_KEY_MAGIC);
 329                 /*get key and mac from request buffer*/
 330         if (valid & ATTR_MAC) {
 331                 memcpy(md_key->md_ck.ck_mac, ckey->ck_mac, MAC_SIZE);
 332                         CDEBUG(D_INFO, "set mac %s for ino %lu \n",
 333                                         md_key->md_ck.ck_mac, inode->i_ino);
 334         }
 335         if (valid & ATTR_KEY) {
 336                 memcpy(md_key->md_ck.ck_key, ckey->ck_key, KEY_SIZE);
 337                        CDEBUG(D_INFO, "set key %s for ino %lu \n",
 338                                         md_key->md_ck.ck_key, inode->i_ino);
 339         }
 340         rc = fsfilt_set_md(obd, inode, handle, md_key, sizeof(*md_key), EA_KEY);
 341 free:
 342         if (md_key)
 343                 OBD_FREE(md_key, sizeof(*md_key));
 344         RETURN(rc);
 345 }
 346
 347 int mds_set_crypto_type(struct obd_device *obd, void *val, __u32 vallen)
 348 {
 349         struct mds_obd *mds = &obd->u.mds;
 350         ENTRY;
 351         if (vallen >= strlen("mks") &&
 352              memcmp(val, "mks", vallen) == 0) {
 353                 mds->mds_crypto_type = MKS_TYPE;
 354                 CDEBUG(D_IOCTL, "mks type\n");
 355         }
 356         if (vallen >= strlen("gks") &&
 357              memcmp(val, "gks", vallen) == 0) {
 358                 mds->mds_crypto_type = GKS_TYPE;
 359                 CDEBUG(D_IOCTL, "gks type \n");
 360         }
 361         RETURN(0);
 362 }
 363
 364 /* Note that we can copy all of the fields, just some will not be "valid" */
 365 void mds_pack_inode2body(struct obd_device *obd, struct mds_body *b,
 366                          struct inode *inode, int fid)
 367 {
 368         b->valid |= OBD_MD_FLID | OBD_MD_FLCTIME | OBD_MD_FLUID |
 369                 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLTYPE |
 370                 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLGENER |
 371                 OBD_MD_FLATIME | OBD_MD_FLMTIME; /* bug 2020 */
 372
 373         if (!S_ISREG(inode->i_mode)) {
 374                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
 375                         OBD_MD_FLATIME | OBD_MD_FLMTIME |
 376                         OBD_MD_FLRDEV;
 377         }
 378         b->atime = LTIME_S(inode->i_atime);
 379         b->mtime = LTIME_S(inode->i_mtime);
 380         b->ctime = LTIME_S(inode->i_ctime);
 381         b->mode = inode->i_mode;
 382         b->size = inode->i_size;
 383         b->blocks = inode->i_blocks;
 384         b->uid = inode->i_uid;
 385         b->gid = inode->i_gid;
 386         b->flags = inode->i_flags;
 387         b->rdev = inode->i_rdev;
 388
 389         /* Return the correct link count for orphan inodes */
 390         if (mds_inode_is_orphan(inode)) {
 391                 b->nlink = 0;
 392         } else if (S_ISDIR(inode->i_mode)) {
 393                 b->nlink = 1;
 394         } else {
 395                 b->nlink = inode->i_nlink;
 396         }
 397
 398         if (fid)
 399                 b->valid |= OBD_MD_FID;
 400
 401         mds_pack_inode2id(obd, &b->id1, inode, fid);
 402 }
 403
 404 /* unpacking */
 405 static int mds_setattr_unpack(struct ptlrpc_request *req, int offset,
 406                               struct mds_update_record *r)
 407 {
 408         struct iattr *attr = &r->ur_iattr;
 409         struct mds_rec_setattr *rec;
 410         ENTRY;
 411
 412         rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
 413                                  lustre_swab_mds_rec_setattr);
 414         if (rec == NULL)
 415                 RETURN (-EFAULT);
 416
 417         r->ur_id1 = &rec->sa_id;
 418         r->ur_flags = rec->sa_flags;
 419         attr->ia_valid = rec->sa_valid;
 420         attr->ia_mode = rec->sa_mode;
 421         attr->ia_uid = rec->sa_uid;
 422         attr->ia_gid = rec->sa_gid;
 423         attr->ia_size = rec->sa_size;
 424         LTIME_S(attr->ia_atime) = rec->sa_atime;
 425         LTIME_S(attr->ia_mtime) = rec->sa_mtime;
 426         LTIME_S(attr->ia_ctime) = rec->sa_ctime;
 427         attr->ia_attr_flags = rec->sa_attr_flags;
 428
 429         LASSERT_REQSWAB(req, offset + 1);
 430         if (req->rq_reqmsg->bufcount > offset + 1) {
 431                 r->ur_eadata = lustre_msg_buf(req->rq_reqmsg,
 432                                               offset + 1, 0);
 433                 if (r->ur_eadata == NULL)
 434                         RETURN (-EFAULT);
 435                 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 1];
 436         }
 437
 438         if (req->rq_reqmsg->bufcount > offset + 2) {
 439                 r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
 440                 if (r->ur_ea2data == NULL)
 441                         RETURN (-EFAULT);
 442
 443                 r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 2];
 444         }
 445
 446         if (req->rq_reqmsg->bufcount > offset + 3) {
 447                 r->ur_ea3data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
 448                 if (r->ur_ea3data == NULL)
 449                         RETURN (-EFAULT);
 450
 451                 r->ur_ea3datalen = req->rq_reqmsg->buflens[offset + 3];
 452         }
 453
 454         RETURN(0);
 455 }
 456
 457 static int mds_create_unpack(struct ptlrpc_request *req, int offset,
 458                              struct mds_update_record *r)
 459 {
 460         struct mds_rec_create *rec;
 461         ENTRY;
 462
 463         rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
 464                                  lustre_swab_mds_rec_create);
 465         if (rec == NULL)
 466                 RETURN (-EFAULT);
 467
 468         r->ur_id1 = &rec->cr_id;
 469         r->ur_id2 = &rec->cr_replayid;
 470         r->ur_mode = rec->cr_mode;
 471         r->ur_rdev = rec->cr_rdev;
 472         r->ur_time = rec->cr_time;
 473         r->ur_flags = rec->cr_flags;
 474
 475         LASSERT_REQSWAB(req, offset + 1);
 476         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 477         if (r->ur_name == NULL)
 478                 RETURN(-EFAULT);
 479         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 480
 481         LASSERT_REQSWAB(req, offset + 2);
 482         if (req->rq_reqmsg->bufcount > offset + 2) {
 483                 if (S_ISLNK(r->ur_mode)) {
 484                         r->ur_tgt = lustre_msg_string(req->rq_reqmsg,
 485                                                       offset + 2, 0);
 486                         if (r->ur_tgt == NULL)
 487                                 RETURN(-EFAULT);
 488                         r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
 489                 } else if (S_ISDIR(r->ur_mode) ) {
 490                         /* Stripe info for mkdir - just a 16bit integer */
 491                         if (req->rq_reqmsg->buflens[offset + 2] != 2) {
 492                                 CERROR("mkdir stripe info does not match "
 493                                        "expected size %d vs 2\n",
 494                                        req->rq_reqmsg->buflens[offset + 2]);
 495                                 RETURN(-EINVAL);
 496                         }
 497                         r->ur_eadata = lustre_swab_buf(req->rq_reqmsg,
 498                                                        offset + 2, 2,
 499                                                        __swab16s);
 500                         r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
 501                 } else if (S_ISREG(r->ur_mode)){
 502                         r->ur_eadata = lustre_msg_buf(req->rq_reqmsg,
 503                                                       offset + 2, 0);
 504                         r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
 505                 } else {
 506                         /* Hm, no other users so far? */
 507                         LBUG();
 508                 }
 509         }
 510         RETURN(0);
 511 }
 512
 513 static int mds_link_unpack(struct ptlrpc_request *req, int offset,
 514                            struct mds_update_record *r)
 515 {
 516         struct mds_rec_link *rec;
 517         ENTRY;
 518
 519         rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
 520                                  lustre_swab_mds_rec_link);
 521         if (rec == NULL)
 522                 RETURN(-EFAULT);
 523
 524         r->ur_id1 = &rec->lk_id1;
 525         r->ur_id2 = &rec->lk_id2;
 526         r->ur_time = rec->lk_time;
 527         r->ur_flags = rec->lk_flags;
 528
 529         LASSERT_REQSWAB(req, offset + 1);
 530         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 531         if (r->ur_name == NULL)
 532                 RETURN(-EFAULT);
 533         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 534         RETURN(0);
 535 }
 536
 537 static int mds_unlink_unpack(struct ptlrpc_request *req, int offset,
 538                              struct mds_update_record *r)
 539 {
 540         struct mds_rec_unlink *rec;
 541         ENTRY;
 542
 543         rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
 544                                  lustre_swab_mds_rec_unlink);
 545         if (rec == NULL)
 546                 RETURN(-EFAULT);
 547
 548         r->ur_mode = rec->ul_mode;
 549         r->ur_id1 = &rec->ul_id1;
 550         r->ur_id2 = &rec->ul_id2;
 551         r->ur_time = rec->ul_time;
 552         r->ur_flags = rec->ul_flags;
 553
 554         LASSERT_REQSWAB(req, offset + 1);
 555         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 556         if (r->ur_name == NULL)
 557                 RETURN(-EFAULT);
 558         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 559         RETURN(0);
 560 }
 561
 562 static int mds_rename_unpack(struct ptlrpc_request *req, int offset,
 563                              struct mds_update_record *r)
 564 {
 565         struct mds_rec_rename *rec;
 566         ENTRY;
 567
 568         rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
 569                                  lustre_swab_mds_rec_rename);
 570         if (rec == NULL)
 571                 RETURN(-EFAULT);
 572
 573         r->ur_id1 = &rec->rn_id1;
 574         r->ur_id2 = &rec->rn_id2;
 575         r->ur_time = rec->rn_time;
 576         r->ur_flags = rec->rn_flags;
 577
 578         LASSERT_REQSWAB(req, offset + 1);
 579         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 580         if (r->ur_name == NULL)
 581                 RETURN(-EFAULT);
 582         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 583
 584         LASSERT_REQSWAB(req, offset + 2);
 585         r->ur_tgt = lustre_msg_string(req->rq_reqmsg, offset + 2, 0);
 586         if (r->ur_tgt == NULL)
 587                 RETURN(-EFAULT);
 588         r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
 589         RETURN(0);
 590 }
 591
 592 static int mds_open_unpack(struct ptlrpc_request *req, int offset,
 593                            struct mds_update_record *r)
 594 {
 595         struct mds_rec_create *rec;
 596         ENTRY;
 597
 598         rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
 599                                  lustre_swab_mds_rec_create);
 600         if (rec == NULL)
 601                 RETURN(-EFAULT);
 602
 603         r->ur_id1 = &rec->cr_id;
 604         r->ur_id2 = &rec->cr_replayid;
 605         r->ur_mode = rec->cr_mode;
 606         r->ur_rdev = rec->cr_rdev;
 607         r->ur_time = rec->cr_time;
 608         r->ur_flags = rec->cr_flags;
 609
 610         LASSERT_REQSWAB(req, offset + 1);
 611         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 612
 613         if (r->ur_name == NULL)
 614                 RETURN(-EFAULT);
 615         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 616
 617         LASSERT_REQSWAB(req, offset + 2);
 618         if (req->rq_reqmsg->bufcount > offset + 2) {
 619                 r->ur_eadata = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
 620                 if (r->ur_eadata == NULL)
 621                         RETURN(-EFAULT);
 622                 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
 623         }
 624
 625         if (rec->cr_flags & MDS_OPEN_HAS_KEY) {
 626                 LASSERT(req->rq_reqmsg->bufcount > offset + 3);
 627                 r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
 628                 r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 3];
 629         }
 630         RETURN(0);
 631 }
 632
 633 typedef int (*update_unpacker)(struct ptlrpc_request *req, int offset,
 634                                struct mds_update_record *r);
 635
 636 static update_unpacker mds_unpackers[REINT_MAX + 1] = {
 637         [REINT_SETATTR] mds_setattr_unpack,
 638         [REINT_CREATE] mds_create_unpack,
 639         [REINT_LINK] mds_link_unpack,
 640         [REINT_UNLINK] mds_unlink_unpack,
 641         [REINT_RENAME] mds_rename_unpack,
 642         [REINT_OPEN] mds_open_unpack,
 643 };
 644
 645 int mds_update_unpack(struct ptlrpc_request *req, int offset,
 646                       struct mds_update_record *rec)
 647 {
 648         __u32 *opcodep;
 649         __u32  opcode;
 650         int rc;
 651         ENTRY;
 652
 653         /* NB don't lustre_swab_reqbuf() here. We're just taking a peek and we
 654          * want to leave it to the specific unpacker once we've identified the
 655          * message type. */
 656         opcodep = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*opcodep));
 657         if (opcodep == NULL)
 658                 RETURN(-EFAULT);
 659
 660         opcode = *opcodep;
 661         if (lustre_msg_swabbed(req->rq_reqmsg))
 662                 __swab32s(&opcode);
 663
 664         if (opcode > REINT_MAX ||
 665             mds_unpackers[opcode] == NULL) {
 666                 CERROR("Unexpected opcode %d\n", opcode);
 667                 RETURN(-EFAULT);
 668         }
 669
 670         rec->ur_id1 = NULL;
 671         rec->ur_id2 = NULL;
 672         rec->ur_opcode = opcode;
 673
 674         rc = mds_unpackers[opcode](req, offset, rec);
 675
 676 #if CRAY_PORTALS
 677         rec->ur_fsuid = req->rq_uid;
 678 #endif
 679         RETURN(rc);
 680 }
 681
 682 /*
 683  * here we take simple rule: once uid/fsuid is root, we also squash
 684  * the gid/fsgid, don't care setuid/setgid attributes.
 685  */
 686 static
 687 int mds_squash_root(struct mds_obd *mds, struct mds_req_sec_desc *rsd,
 688                     ptl_nid_t *peernid)
 689 {
 690         if (!mds->mds_squash_uid || *peernid == mds->mds_nosquash_nid)
 691                 return 0;
 692
 693         if (rsd->rsd_uid && rsd->rsd_fsuid)
 694                 return 0;
 695
 696         CDEBUG(D_SEC, "squash req from "LPX64":"
 697                "(%u:%u-%u:%u/%x)=>(%u:%u-%u:%u/%x)\n", *peernid,
 698                 rsd->rsd_uid, rsd->rsd_gid,
 699                 rsd->rsd_fsuid, rsd->rsd_fsgid, rsd->rsd_cap,
 700                 rsd->rsd_uid ? rsd->rsd_uid : mds->mds_squash_uid,
 701                 rsd->rsd_uid ? rsd->rsd_gid : mds->mds_squash_gid,
 702                 rsd->rsd_fsuid ? rsd->rsd_fsuid : mds->mds_squash_uid,
 703                 rsd->rsd_fsuid ? rsd->rsd_fsgid : mds->mds_squash_gid,
 704                 rsd->rsd_cap & ~CAP_FS_MASK);
 705
 706         if (rsd->rsd_uid == 0) {
 707                 rsd->rsd_uid = mds->mds_squash_uid;
 708                 rsd->rsd_gid = mds->mds_squash_gid;
 709         }
 710         if (rsd->rsd_fsuid == 0) {
 711                 rsd->rsd_fsuid = mds->mds_squash_uid;
 712                 rsd->rsd_fsgid = mds->mds_squash_gid;
 713         }
 714         rsd->rsd_cap &= ~CAP_FS_MASK;
 715
 716         return 1;
 717 }
 718
 719 /********************************
 720  * MDS uid/gid mapping handling *
 721  ********************************/
 722
 723 static
 724 struct mds_idmap_entry* idmap_alloc_entry(__u32 rmt_id, __u32 lcl_id)
 725 {
 726         struct mds_idmap_entry *e;
 727
 728         OBD_ALLOC(e, sizeof(*e));
 729         if (!e)
 730                 return NULL;
 731
 732         INIT_LIST_HEAD(&e->rmt_hash);
 733         INIT_LIST_HEAD(&e->lcl_hash);
 734         atomic_set(&e->refcount, 1);
 735         e->rmt_id = rmt_id;
 736         e->lcl_id = lcl_id;
 737
 738         return e;
 739 }
 740
 741 void idmap_free_entry(struct mds_idmap_entry *e)
 742 {
 743         if (!list_empty(&e->rmt_hash))
 744                 list_del(&e->rmt_hash);
 745         if (!list_empty(&e->lcl_hash))
 746                 list_del(&e->lcl_hash);
 747         OBD_FREE(e, sizeof(*e));
 748 }
 749
 750 static
 751 int idmap_insert_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
 752                        struct mds_idmap_entry *new, const char *warn_msg)
 753 {
 754         struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(new->rmt_id)];
 755         struct list_head *lcl_head = &lcl_hash[MDS_IDMAP_HASHFUNC(new->lcl_id)];
 756         struct mds_idmap_entry *e;
 757
 758         list_for_each_entry(e, rmt_head, rmt_hash) {
 759                 if (e->rmt_id == new->rmt_id &&
 760                     e->lcl_id == new->lcl_id) {
 761                         atomic_inc(&e->refcount);
 762                         return 1;
 763                 }
 764                 if (e->rmt_id == new->rmt_id && warn_msg)
 765                         CWARN("%s: rmt id %u already map to %u (new %u)\n",
 766                               warn_msg, e->rmt_id, e->lcl_id, new->lcl_id);
 767                 if (e->lcl_id == new->lcl_id && warn_msg)
 768                         CWARN("%s: lcl id %u already be mapped from %u "
 769                               "(new %u)\n", warn_msg,
 770                               e->lcl_id, e->rmt_id, new->rmt_id);
 771         }
 772
 773         list_add_tail(rmt_head, &new->rmt_hash);
 774         list_add_tail(lcl_head, &new->lcl_hash);
 775         return 0;
 776 }
 777
 778 static
 779 int idmap_remove_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
 780                        __u32 rmt_id, __u32 lcl_id)
 781 {
 782         struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(rmt_id)];
 783         struct mds_idmap_entry *e;
 784
 785         list_for_each_entry(e, rmt_head, rmt_hash) {
 786                 if (e->rmt_id == rmt_id && e->lcl_id == lcl_id) {
 787                         if (atomic_dec_and_test(&e->refcount)) {
 788                                 list_del(&e->rmt_hash);
 789                                 list_del(&e->lcl_hash);
 790                                 OBD_FREE(e, sizeof(*e));
 791                                 return 0;
 792                         } else
 793                                 return 1;
 794                 }
 795         }
 796         return -ENOENT;
 797 }
 798
 799 int mds_idmap_add(struct mds_idmap_table *tbl,
 800                   uid_t rmt_uid, uid_t lcl_uid,
 801                   gid_t rmt_gid, gid_t lcl_gid)
 802 {
 803         struct mds_idmap_entry *ue, *ge;
 804         ENTRY;
 805
 806         if (!tbl)
 807                 RETURN(-EPERM);
 808
 809         ue = idmap_alloc_entry(rmt_uid, lcl_uid);
 810         if (!ue)
 811                 RETURN(-ENOMEM);
 812         ge = idmap_alloc_entry(rmt_gid, lcl_gid);
 813         if (!ge) {
 814                 idmap_free_entry(ue);
 815                 RETURN(-ENOMEM);
 816         }
 817
 818         spin_lock(&tbl->mit_lock);
 819
 820         if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
 821                                tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
 822                                ue, "UID mapping")) {
 823                 idmap_free_entry(ue);
 824         }
 825
 826         if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
 827                                tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
 828                                ge, "GID mapping")) {
 829                 idmap_free_entry(ge);
 830         }
 831
 832         spin_unlock(&tbl->mit_lock);
 833         RETURN(0);
 834 }
 835
 836 int mds_idmap_del(struct mds_idmap_table *tbl,
 837                   uid_t rmt_uid, uid_t lcl_uid,
 838                   gid_t rmt_gid, gid_t lcl_gid)
 839 {
 840         ENTRY;
 841
 842         if (!tbl)
 843                 RETURN(0);
 844
 845         spin_lock(&tbl->mit_lock);
 846         idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
 847                            tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
 848                            rmt_uid, lcl_uid);
 849         idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
 850                            tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
 851                            rmt_gid, lcl_gid);
 852         spin_unlock(&tbl->mit_lock);
 853         RETURN(0);
 854 }
 855
 856 static
 857 __u32 idmap_lookup_id(struct list_head *hash, int reverse, __u32 id)
 858 {
 859         struct list_head *head = &hash[MDS_IDMAP_HASHFUNC(id)];
 860         struct mds_idmap_entry *e;
 861
 862         if (!reverse) {
 863                 list_for_each_entry(e, head, rmt_hash) {
 864                         if (e->rmt_id == id)
 865                                 return e->lcl_id;
 866                 }
 867                 return MDS_IDMAP_NOTFOUND;
 868         } else {
 869                 list_for_each_entry(e, head, lcl_hash) {
 870                         if (e->lcl_id == id)
 871                                 return e->rmt_id;
 872                 }
 873                 return MDS_IDMAP_NOTFOUND;
 874         }
 875 }
 876
 877 int mds_idmap_lookup_uid(struct mds_idmap_table *tbl, int reverse, uid_t uid)
 878 {
 879         struct list_head *hash;
 880
 881         if (!tbl)
 882                 return MDS_IDMAP_NOTFOUND;
 883
 884         if (!reverse)
 885                 hash = tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX];
 886         else
 887                 hash = tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX];
 888
 889         spin_lock(&tbl->mit_lock);
 890         uid = idmap_lookup_id(hash, reverse, uid);
 891         spin_unlock(&tbl->mit_lock);
 892
 893         return uid;
 894 }
 895
 896 int mds_idmap_lookup_gid(struct mds_idmap_table *tbl, int reverse, gid_t gid)
 897 {
 898         struct list_head *hash;
 899
 900         if (!tbl)
 901                 return MDS_IDMAP_NOTFOUND;
 902
 903         if (!reverse)
 904                 hash = tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX];
 905         else
 906                 hash = tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX];
 907
 908         spin_lock(&tbl->mit_lock);
 909         gid = idmap_lookup_id(hash, reverse, gid);
 910         spin_unlock(&tbl->mit_lock);
 911
 912         return gid;
 913 }
 914
 915 struct mds_idmap_table *mds_idmap_alloc()
 916 {
 917         struct mds_idmap_table *tbl;
 918         int i, j;
 919
 920         OBD_ALLOC(tbl, sizeof(*tbl));
 921         if (!tbl)
 922                 return NULL;
 923
 924         spin_lock_init(&tbl->mit_lock);
 925         for (i = 0; i < MDS_IDMAP_N_HASHES; i++)
 926                 for (j = 0; j < MDS_IDMAP_HASHSIZE; j++)
 927                         INIT_LIST_HEAD(&tbl->mit_idmaps[i][j]);
 928
 929         return tbl;
 930 }
 931
 932 static void idmap_clear_rmt_hash(struct list_head *list)
 933 {
 934         struct mds_idmap_entry *e;
 935         int i;
 936
 937         for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
 938                 while (!list_empty(&list[i])) {
 939                         e = list_entry(list[i].next, struct mds_idmap_entry,
 940                                        rmt_hash);
 941                         idmap_free_entry(e);
 942                 }
 943         }
 944 }
 945
 946 void mds_idmap_free(struct mds_idmap_table *tbl)
 947 {
 948         int i;
 949
 950         spin_lock(&tbl->mit_lock);
 951         idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX]);
 952         idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX]);
 953
 954         /* paranoid checking */
 955         for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
 956                 LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX][i]));
 957                 LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX][i]));
 958         }
 959         spin_unlock(&tbl->mit_lock);
 960
 961         OBD_FREE(tbl, sizeof(*tbl));
 962 }
 963
 964 /*********************************
 965  * helpers doing mapping for MDS *
 966  *********************************/
 967
 968 /*
 969  * we allow remote setuid/setgid to an "authencated" one,
 970  * this policy probably change later.
 971  */
 972 static
 973 int mds_req_secdesc_do_map(struct mds_export_data *med,
 974                            struct mds_req_sec_desc *rsd)
 975 {
 976         struct mds_idmap_table *idmap = med->med_idmap;
 977         uid_t uid, fsuid;
 978         gid_t gid, fsgid;
 979
 980         uid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_uid);
 981         if (uid == MDS_IDMAP_NOTFOUND) {
 982                 CERROR("can't find map for uid %u\n", rsd->rsd_uid);
 983                 return -EPERM;
 984         }
 985
 986         if (rsd->rsd_uid == rsd->rsd_fsuid)
 987                 fsuid = uid;
 988         else {
 989                 fsuid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_fsuid);
 990                 if (fsuid == MDS_IDMAP_NOTFOUND) {
 991                         CERROR("can't find map for fsuid %u\n", rsd->rsd_fsuid);
 992                         return -EPERM;
 993                 }
 994         }
 995
 996         gid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_gid);
 997         if (gid == MDS_IDMAP_NOTFOUND) {
 998                 CERROR("can't find map for gid %u\n", rsd->rsd_gid);
 999                 return -EPERM;
1000         }
1001
1002         if (rsd->rsd_gid == rsd->rsd_fsgid)
1003                 fsgid = gid;
1004         else {
1005                 fsgid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_fsgid);
1006                 if (fsgid == MDS_IDMAP_NOTFOUND) {
1007                         CERROR("can't find map for fsgid %u\n", rsd->rsd_fsgid);
1008                         return -EPERM;
1009                 }
1010         }
1011
1012         rsd->rsd_uid = uid;
1013         rsd->rsd_gid = gid;
1014         rsd->rsd_fsuid = fsuid;
1015         rsd->rsd_fsgid = fsgid;
1016
1017         return 0;
1018 }
1019
1020 void mds_body_do_reverse_map(struct mds_export_data *med,
1021                              struct mds_body *body)
1022 {
1023         uid_t uid;
1024         gid_t gid;
1025
1026         if (!med->med_remote)
1027                 return;
1028
1029         ENTRY;
1030         if (body->valid & OBD_MD_FLUID) {
1031                 uid = mds_idmap_lookup_uid(med->med_idmap, 1, body->uid);
1032                 if (uid == MDS_IDMAP_NOTFOUND) {
1033                         uid = med->med_nllu;
1034                         if (body->valid & OBD_MD_FLMODE) {
1035                                 body->mode = (body->mode & ~S_IRWXU) |
1036                                              ((body->mode & S_IRWXO) << 6);
1037                         }
1038                 }
1039                 body->uid = uid;
1040         }
1041         if (body->valid & OBD_MD_FLGID) {
1042                 gid = mds_idmap_lookup_gid(med->med_idmap, 1, body->gid);
1043                 if (gid == MDS_IDMAP_NOTFOUND) {
1044                         gid = med->med_nllg;
1045                         if (body->valid & OBD_MD_FLMODE) {
1046                                 body->mode = (body->mode & ~S_IRWXG) |
1047                                              ((body->mode & S_IRWXO) << 3);
1048                         }
1049                 }
1050                 body->gid = gid;
1051         }
1052
1053         EXIT;
1054 }
1055
1056 /**********************
1057  * MDS ucred handling *
1058  **********************/
1059
1060 static inline void drop_ucred_ginfo(struct lvfs_ucred *ucred)
1061 {
1062         if (ucred->luc_ginfo) {
1063                 put_group_info(ucred->luc_ginfo);
1064                 ucred->luc_ginfo = NULL;
1065         }
1066 }
1067
1068 static inline void drop_ucred_lsd(struct lvfs_ucred *ucred)
1069 {
1070         if (ucred->luc_lsd) {
1071                 mds_put_lsd(ucred->luc_lsd);
1072                 ucred->luc_lsd = NULL;
1073         }
1074 }
1075
1076 /*
1077  * the heart of the uid/gid handling and security checking.
1078  *
1079  * root could set any group_info if we allowed setgroups, while
1080  * normal user only could 'reduce' their group members -- which
1081  * is somewhat expensive.
1082  *
1083  * authenticated as mds user (using mds service credential) could
1084  * bypass all checkings.
1085  */
1086 int mds_init_ucred(struct lvfs_ucred *ucred,
1087                    struct ptlrpc_request *req,
1088                    struct mds_req_sec_desc *rsd)
1089 {
1090         struct mds_obd *mds = &req->rq_export->exp_obd->u.mds;
1091         struct mds_export_data *med = &req->rq_export->u.eu_mds_data;
1092         struct lustre_sec_desc *lsd;
1093         ptl_nid_t peernid = req->rq_peer.peer_id.nid;
1094         struct group_info *gnew;
1095         unsigned int setuid, setgid, strong_sec, root_squashed;
1096         __u32 lsd_perms;
1097         ENTRY;
1098
1099         LASSERT(ucred);
1100         LASSERT(rsd);
1101         LASSERT(rsd->rsd_ngroups <= LUSTRE_MAX_GROUPS);
1102
1103         if (SEC_FLAVOR_MAJOR(req->rq_req_secflvr) == PTLRPCS_FLVR_MAJOR_GSS &&
1104             (SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_AUTH ||
1105              SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_PRIV))
1106                 strong_sec = 1;
1107         else
1108                 strong_sec = 0;
1109
1110         LASSERT(!(req->rq_remote_realm && !strong_sec));
1111
1112         if (strong_sec && req->rq_auth_uid == -1) {
1113                 CWARN("user not authenticated, deny access\n");
1114                 RETURN(-EPERM);
1115         }
1116
1117         /* sanity check: if we use strong authentication, we expect the
1118          * uid which client claimed is true.
1119          * not apply to special mds user .
1120          */
1121         if (!req->rq_auth_usr_mds && strong_sec) {
1122                 if (!med->med_remote) {
1123                         if (req->rq_auth_uid != rsd->rsd_uid) {
1124                                 CERROR("local client "LPU64": auth uid %u "
1125                                        "while client claim %u:%u/%u:%u\n",
1126                                        peernid, req->rq_auth_uid,
1127                                        rsd->rsd_uid, rsd->rsd_gid,
1128                                        rsd->rsd_fsuid, rsd->rsd_fsgid);
1129                                 RETURN(-EPERM);
1130                         }
1131                 } else {
1132                         if (req->rq_mapped_uid == MDS_IDMAP_NOTFOUND) {
1133                                 CWARN("no mapping found, deny\n");
1134                                 RETURN(-EPERM);
1135                         }
1136
1137                         if (mds_req_secdesc_do_map(med, rsd))
1138                                 RETURN(-EPERM);
1139
1140                         if (req->rq_mapped_uid != rsd->rsd_uid) {
1141                                 CERROR("remote client "LPU64": auth uid %u "
1142                                        "while client claim %u:%u/%u:%u\n",
1143                                        peernid, req->rq_auth_uid,
1144                                        rsd->rsd_uid, rsd->rsd_gid,
1145                                        rsd->rsd_fsuid, rsd->rsd_fsgid);
1146                                 RETURN(-EPERM);
1147                         }
1148                 }
1149         }
1150
1151         /* now LSD come into play */
1152         ucred->luc_ginfo = NULL;
1153         ucred->luc_lsd = lsd = mds_get_lsd(rsd->rsd_uid);
1154
1155         if (!lsd) {
1156                 CERROR("Deny access without LSD: uid %d\n", rsd->rsd_uid);
1157                 RETURN(-EPERM);
1158         }
1159
1160         lsd_perms = mds_lsd_get_perms(lsd, med->med_remote, 0, peernid);
1161
1162         /* check setuid/setgid permissions.
1163          * again not apply to special mds user.
1164          */
1165         if (!req->rq_auth_usr_mds) {
1166                 /* find out the setuid/setgid attempt */
1167                 setuid = (rsd->rsd_uid != rsd->rsd_fsuid);
1168                 setgid = (rsd->rsd_gid != rsd->rsd_fsgid ||
1169                           rsd->rsd_gid != lsd->lsd_gid);
1170
1171                 /* check permission of setuid */
1172                 if (setuid && !(lsd_perms & LSD_PERM_SETUID)) {
1173                         CWARN("mds blocked setuid attempt (%u -> %u) "
1174                               "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_fsuid,
1175                               peernid);
1176                         RETURN(-EPERM);
1177                 }
1178
1179                 /* check permission of setgid */
1180                 if (setgid && !(lsd_perms & LSD_PERM_SETGID)) {
1181                         CWARN("mds blocked setgid attempt (%u:%u/%u:%u -> %u) "
1182                               "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_gid,
1183                               rsd->rsd_fsuid, rsd->rsd_fsgid, lsd->lsd_gid,
1184                               peernid);
1185                         RETURN(-EPERM);
1186                 }
1187         }
1188
1189         root_squashed = mds_squash_root(mds, rsd, &peernid);
1190
1191         /* remove privilege for non-root user */
1192         if (rsd->rsd_fsuid)
1193                 rsd->rsd_cap &= ~CAP_FS_MASK;
1194
1195         /* by now every fields other than groups in rsd have been granted */
1196         ucred->luc_nid = peernid;
1197         ucred->luc_uid = rsd->rsd_uid;
1198         ucred->luc_gid = rsd->rsd_gid;
1199         ucred->luc_fsuid = rsd->rsd_fsuid;
1200         ucred->luc_fsgid = rsd->rsd_fsgid;
1201         ucred->luc_cap = rsd->rsd_cap;
1202
1203         /* don't use any supplementary group if we squashed root.
1204          * XXX The exact behavior of root_squash is not defined, we just
1205          * keep the reminder here */
1206         if (root_squashed)
1207                 RETURN(0);
1208
1209         /* install groups from LSD */
1210         if (lsd->lsd_ginfo) {
1211                 ucred->luc_ginfo = lsd->lsd_ginfo;
1212                 get_group_info(ucred->luc_ginfo);
1213         }
1214
1215         /* everything is done if we don't allow setgroups, or it is
1216          * from remote client (which implies forced to be no-setgroups).
1217          *
1218          * Note: remote user's supplementary groups sent along the request
1219          * (if any) are all ignored, but we make the mapped local user's
1220          * supplementary groups take effect.
1221          */
1222         if (med->med_remote || !(lsd_perms & LSD_PERM_SETGRP))
1223                 RETURN(0);
1224
1225         /* root could set any groups as he want (if allowed), normal
1226          * users only could reduce his group array.
1227          */
1228         if (ucred->luc_uid == 0) {
1229                 drop_ucred_ginfo(ucred);
1230
1231                 if (rsd->rsd_ngroups == 0)
1232                         RETURN(0);
1233
1234                 gnew = groups_alloc(rsd->rsd_ngroups);
1235                 if (!gnew) {
1236                         CERROR("out of memory\n");
1237                         drop_ucred_lsd(ucred);
1238                         RETURN(-ENOMEM);
1239                 }
1240                 groups_from_buffer(gnew, rsd->rsd_groups);
1241                 groups_sort(gnew); /* don't rely on client doing this */
1242
1243                 ucred->luc_ginfo = gnew;
1244         } else {
1245                 __u32 set = 0, cur = 0;
1246                 struct group_info *ginfo = ucred->luc_ginfo;
1247
1248                 if (!ginfo)
1249                         RETURN(0);
1250
1251                 /* Note: freeing a group_info count on 'nblocks' instead of
1252                  * 'ngroups', thus we can safely alloc enough buffer and reduce
1253                  * and ngroups number later.
1254                  */
1255                 gnew = groups_alloc(rsd->rsd_ngroups);
1256                 if (!gnew) {
1257                         CERROR("out of memory\n");
1258                         drop_ucred_ginfo(ucred);
1259                         drop_ucred_lsd(ucred);
1260                         RETURN(-ENOMEM);
1261                 }
1262
1263                 while (cur < rsd->rsd_ngroups) {
1264                         if (groups_search(ginfo, rsd->rsd_groups[cur])) {
1265                                 GROUP_AT(gnew, set) = rsd->rsd_groups[cur];
1266                                 set++;
1267                         }
1268                         cur++;
1269                 }
1270                 gnew->ngroups = set;
1271
1272                 put_group_info(ucred->luc_ginfo);
1273                 ucred->luc_ginfo = gnew;
1274         }
1275         RETURN(0);
1276 }
1277
1278 void mds_exit_ucred(struct lvfs_ucred *ucred)
1279 {
1280         ENTRY;
1281         drop_ucred_ginfo(ucred);
1282         drop_ucred_lsd(ucred);
1283         EXIT;
1284 }
1285