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