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         attr->ia_valid = rec->sa_valid;
 426         attr->ia_mode = rec->sa_mode;
 427         attr->ia_uid = rec->sa_uid;
 428         attr->ia_gid = rec->sa_gid;
 429         attr->ia_size = rec->sa_size;
 430         LTIME_S(attr->ia_atime) = rec->sa_atime;
 431         LTIME_S(attr->ia_mtime) = rec->sa_mtime;
 432         LTIME_S(attr->ia_ctime) = rec->sa_ctime;
 433         attr->ia_attr_flags = rec->sa_attr_flags;
 434
 435         LASSERT_REQSWAB (req, offset + 1);
 436         if (req->rq_reqmsg->bufcount > offset + 1) {
 437                 r->ur_eadata = lustre_msg_buf (req->rq_reqmsg,
 438                                                offset + 1, 0);
 439                 if (r->ur_eadata == NULL)
 440                         RETURN (-EFAULT);
 441                 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 1];
 442         }
 443
 444         if (req->rq_reqmsg->bufcount > offset + 2) {
 445                 r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
 446                 if (r->ur_ea2data == NULL)
 447                         RETURN (-EFAULT);
 448
 449                 r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 2];
 450         }
 451
 452         if (req->rq_reqmsg->bufcount > offset + 3) {
 453                 r->ur_ea3data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
 454                 if (r->ur_ea3data == NULL)
 455                         RETURN (-EFAULT);
 456
 457                 r->ur_ea3datalen = req->rq_reqmsg->buflens[offset + 3];
 458         }
 459
 460         RETURN(0);
 461 }
 462
 463 static int mds_create_unpack(struct ptlrpc_request *req, int offset,
 464                              struct mds_update_record *r)
 465 {
 466         struct mds_rec_create *rec;
 467         ENTRY;
 468
 469         rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
 470                                   lustre_swab_mds_rec_create);
 471         if (rec == NULL)
 472                 RETURN (-EFAULT);
 473
 474         r->ur_id1 = &rec->cr_id;
 475         r->ur_id2 = &rec->cr_replayid;
 476         r->ur_mode = rec->cr_mode;
 477         r->ur_rdev = rec->cr_rdev;
 478         r->ur_time = rec->cr_time;
 479         r->ur_flags = rec->cr_flags;
 480
 481         LASSERT_REQSWAB (req, offset + 1);
 482         r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
 483         if (r->ur_name == NULL)
 484                 RETURN (-EFAULT);
 485         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 486
 487         LASSERT_REQSWAB (req, offset + 2);
 488         if (req->rq_reqmsg->bufcount > offset + 2) {
 489                 if (S_ISLNK(r->ur_mode)) {
 490                         r->ur_tgt = lustre_msg_string(req->rq_reqmsg,
 491                                                       offset + 2, 0);
 492                         if (r->ur_tgt == NULL)
 493                                 RETURN (-EFAULT);
 494                         r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
 495                 } else if (S_ISDIR(r->ur_mode) ) {
 496                         /* Stripe info for mkdir - just a 16bit integer */
 497                         if (req->rq_reqmsg->buflens[offset + 2] != 2) {
 498                                 CERROR("mkdir stripe info does not match "
 499                                        "expected size %d vs 2\n",
 500                                        req->rq_reqmsg->buflens[offset + 2]);
 501                                 RETURN (-EINVAL);
 502                         }
 503                         r->ur_eadata = lustre_swab_buf (req->rq_reqmsg,
 504                                                offset + 2, 2, __swab16s);
 505                         r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
 506                 } else if (S_ISREG(r->ur_mode)){
 507                         r->ur_eadata = lustre_msg_buf (req->rq_reqmsg,
 508                                                        offset + 2, 0);
 509                         r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
 510                 } else {
 511                         /* Hm, no other users so far? */
 512                         LBUG();
 513                 }
 514         }
 515         RETURN(0);
 516 }
 517
 518 static int mds_link_unpack(struct ptlrpc_request *req, int offset,
 519                            struct mds_update_record *r)
 520 {
 521         struct mds_rec_link *rec;
 522         ENTRY;
 523
 524         rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
 525                                   lustre_swab_mds_rec_link);
 526         if (rec == NULL)
 527                 RETURN (-EFAULT);
 528
 529         r->ur_id1 = &rec->lk_id1;
 530         r->ur_id2 = &rec->lk_id2;
 531         r->ur_time = rec->lk_time;
 532
 533         LASSERT_REQSWAB (req, offset + 1);
 534         r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
 535         if (r->ur_name == NULL)
 536                 RETURN (-EFAULT);
 537         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 538         RETURN(0);
 539 }
 540
 541 static int mds_unlink_unpack(struct ptlrpc_request *req, int offset,
 542                              struct mds_update_record *r)
 543 {
 544         struct mds_rec_unlink *rec;
 545         ENTRY;
 546
 547         rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
 548                                   lustre_swab_mds_rec_unlink);
 549         if (rec == NULL)
 550                 RETURN(-EFAULT);
 551
 552         r->ur_mode = rec->ul_mode;
 553         r->ur_id1 = &rec->ul_id1;
 554         r->ur_id2 = &rec->ul_id2;
 555         r->ur_time = rec->ul_time;
 556
 557         LASSERT_REQSWAB (req, offset + 1);
 558         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 559         if (r->ur_name == NULL)
 560                 RETURN(-EFAULT);
 561         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 562         RETURN(0);
 563 }
 564
 565 static int mds_rename_unpack(struct ptlrpc_request *req, int offset,
 566                              struct mds_update_record *r)
 567 {
 568         struct mds_rec_rename *rec;
 569         ENTRY;
 570
 571         rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
 572                                   lustre_swab_mds_rec_rename);
 573         if (rec == NULL)
 574                 RETURN(-EFAULT);
 575
 576         r->ur_id1 = &rec->rn_id1;
 577         r->ur_id2 = &rec->rn_id2;
 578         r->ur_time = rec->rn_time;
 579
 580         LASSERT_REQSWAB (req, offset + 1);
 581         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 582         if (r->ur_name == NULL)
 583                 RETURN(-EFAULT);
 584         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 585
 586         LASSERT_REQSWAB (req, offset + 2);
 587         r->ur_tgt = lustre_msg_string(req->rq_reqmsg, offset + 2, 0);
 588         if (r->ur_tgt == NULL)
 589                 RETURN(-EFAULT);
 590         r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
 591         RETURN(0);
 592 }
 593
 594 static int mds_open_unpack(struct ptlrpc_request *req, int offset,
 595                            struct mds_update_record *r)
 596 {
 597         struct mds_rec_create *rec;
 598         ENTRY;
 599
 600         rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
 601                                   lustre_swab_mds_rec_create);
 602         if (rec == NULL)
 603                 RETURN(-EFAULT);
 604
 605         r->ur_id1 = &rec->cr_id;
 606         r->ur_id2 = &rec->cr_replayid;
 607         r->ur_mode = rec->cr_mode;
 608         r->ur_rdev = rec->cr_rdev;
 609         r->ur_time = rec->cr_time;
 610         r->ur_flags = rec->cr_flags;
 611
 612         LASSERT_REQSWAB (req, offset + 1);
 613         r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
 614         if (r->ur_name == NULL)
 615                 RETURN (-EFAULT);
 616         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 617
 618         LASSERT_REQSWAB (req, offset + 2);
 619
 620         if (req->rq_reqmsg->bufcount > offset + 2) {
 621                 r->ur_eadata = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
 622                 if (r->ur_eadata == NULL)
 623                         RETURN(-EFAULT);
 624                 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
 625         }
 626
 627         if (rec->cr_flags & MDS_OPEN_HAS_KEY) {
 628                 LASSERT(req->rq_reqmsg->bufcount > offset + 3);
 629                 r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
 630                 r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 3];
 631         }
 632         RETURN(0);
 633 }
 634
 635 typedef int (*update_unpacker)(struct ptlrpc_request *req, int offset,
 636                                struct mds_update_record *r);
 637
 638 static update_unpacker mds_unpackers[REINT_MAX + 1] = {
 639         [REINT_SETATTR] mds_setattr_unpack,
 640         [REINT_CREATE] mds_create_unpack,
 641         [REINT_LINK] mds_link_unpack,
 642         [REINT_UNLINK] mds_unlink_unpack,
 643         [REINT_RENAME] mds_rename_unpack,
 644         [REINT_OPEN] mds_open_unpack,
 645 };
 646
 647 int mds_update_unpack(struct ptlrpc_request *req, int offset,
 648                       struct mds_update_record *rec)
 649 {
 650         __u32 *opcodep;
 651         __u32  opcode;
 652         int rc;
 653         ENTRY;
 654
 655         /*
 656          * NB don't lustre_swab_reqbuf() here. We're just taking a peek and we
 657          * want to leave it to the specific unpacker once we've identified the
 658          * message type.
 659          */
 660         opcodep = lustre_msg_buf (req->rq_reqmsg, offset, sizeof(*opcodep));
 661         if (opcodep == NULL)
 662                 RETURN(-EFAULT);
 663
 664         opcode = *opcodep;
 665         if (lustre_msg_swabbed (req->rq_reqmsg))
 666                 __swab32s (&opcode);
 667
 668         if (opcode > REINT_MAX ||
 669             mds_unpackers[opcode] == NULL) {
 670                 CERROR ("Unexpected opcode %d\n", opcode);
 671                 RETURN(-EFAULT);
 672         }
 673
 674         rec->ur_id1 = NULL;
 675         rec->ur_id2 = NULL;
 676         rec->ur_opcode = opcode;
 677
 678         rc = mds_unpackers[opcode](req, offset, rec);
 679
 680 #if CRAY_PORTALS
 681         rec->ur_fsuid = req->rq_uid;
 682 #endif
 683         RETURN(rc);
 684 }
 685
 686 /*
 687  * here we take simple rule: once uid/fsuid is root, we also squash
 688  * the gid/fsgid, don't care setuid/setgid attributes.
 689  */
 690 static
 691 int mds_squash_root(struct mds_obd *mds, struct mds_req_sec_desc *rsd,
 692                     ptl_nid_t *peernid)
 693 {
 694         if (!mds->mds_squash_uid || *peernid == mds->mds_nosquash_nid)
 695                 return 0;
 696
 697         if (rsd->rsd_uid && rsd->rsd_fsuid)
 698                 return 0;
 699
 700         CDEBUG(D_SEC, "squash req from "LPX64":"
 701                "(%u:%u-%u:%u/%x)=>(%u:%u-%u:%u/%x)\n", *peernid,
 702                 rsd->rsd_uid, rsd->rsd_gid,
 703                 rsd->rsd_fsuid, rsd->rsd_fsgid, rsd->rsd_cap,
 704                 rsd->rsd_uid ? rsd->rsd_uid : mds->mds_squash_uid,
 705                 rsd->rsd_uid ? rsd->rsd_gid : mds->mds_squash_gid,
 706                 rsd->rsd_fsuid ? rsd->rsd_fsuid : mds->mds_squash_uid,
 707                 rsd->rsd_fsuid ? rsd->rsd_fsgid : mds->mds_squash_gid,
 708                 rsd->rsd_cap & ~CAP_FS_MASK);
 709
 710         if (rsd->rsd_uid == 0) {
 711                 rsd->rsd_uid = mds->mds_squash_uid;
 712                 rsd->rsd_gid = mds->mds_squash_gid;
 713         }
 714         if (rsd->rsd_fsuid == 0) {
 715                 rsd->rsd_fsuid = mds->mds_squash_uid;
 716                 rsd->rsd_fsgid = mds->mds_squash_gid;
 717         }
 718         rsd->rsd_cap &= ~CAP_FS_MASK;
 719
 720         return 1;
 721 }
 722
 723 /********************************
 724  * MDS uid/gid mapping handling *
 725  ********************************/
 726
 727 static
 728 struct mds_idmap_entry* idmap_alloc_entry(__u32 rmt_id, __u32 lcl_id)
 729 {
 730         struct mds_idmap_entry *e;
 731
 732         OBD_ALLOC(e, sizeof(*e));
 733         if (!e)
 734                 return NULL;
 735
 736         INIT_LIST_HEAD(&e->rmt_hash);
 737         INIT_LIST_HEAD(&e->lcl_hash);
 738         atomic_set(&e->refcount, 1);
 739         e->rmt_id = rmt_id;
 740         e->lcl_id = lcl_id;
 741
 742         return e;
 743 }
 744
 745 void idmap_free_entry(struct mds_idmap_entry *e)
 746 {
 747         if (!list_empty(&e->rmt_hash))
 748                 list_del(&e->rmt_hash);
 749         if (!list_empty(&e->lcl_hash))
 750                 list_del(&e->lcl_hash);
 751         OBD_FREE(e, sizeof(*e));
 752 }
 753
 754 static
 755 int idmap_insert_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
 756                        struct mds_idmap_entry *new, const char *warn_msg)
 757 {
 758         struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(new->rmt_id)];
 759         struct list_head *lcl_head = &lcl_hash[MDS_IDMAP_HASHFUNC(new->lcl_id)];
 760         struct mds_idmap_entry *e;
 761
 762         list_for_each_entry(e, rmt_head, rmt_hash) {
 763                 if (e->rmt_id == new->rmt_id &&
 764                     e->lcl_id == new->lcl_id) {
 765                         atomic_inc(&e->refcount);
 766                         return 1;
 767                 }
 768                 if (e->rmt_id == new->rmt_id && warn_msg)
 769                         CWARN("%s: rmt id %u already map to %u (new %u)\n",
 770                               warn_msg, e->rmt_id, e->lcl_id, new->lcl_id);
 771                 if (e->lcl_id == new->lcl_id && warn_msg)
 772                         CWARN("%s: lcl id %u already be mapped from %u "
 773                               "(new %u)\n", warn_msg,
 774                               e->lcl_id, e->rmt_id, new->rmt_id);
 775         }
 776
 777         list_add_tail(rmt_head, &new->rmt_hash);
 778         list_add_tail(lcl_head, &new->lcl_hash);
 779         return 0;
 780 }
 781
 782 static
 783 int idmap_remove_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
 784                        __u32 rmt_id, __u32 lcl_id)
 785 {
 786         struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(rmt_id)];
 787         struct mds_idmap_entry *e;
 788
 789         list_for_each_entry(e, rmt_head, rmt_hash) {
 790                 if (e->rmt_id == rmt_id && e->lcl_id == lcl_id) {
 791                         if (atomic_dec_and_test(&e->refcount)) {
 792                                 list_del(&e->rmt_hash);
 793                                 list_del(&e->lcl_hash);
 794                                 OBD_FREE(e, sizeof(*e));
 795                                 return 0;
 796                         } else
 797                                 return 1;
 798                 }
 799         }
 800         return -ENOENT;
 801 }
 802
 803 int mds_idmap_add(struct mds_idmap_table *tbl,
 804                   uid_t rmt_uid, uid_t lcl_uid,
 805                   gid_t rmt_gid, gid_t lcl_gid)
 806 {
 807         struct mds_idmap_entry *ue, *ge;
 808         ENTRY;
 809
 810         if (!tbl)
 811                 RETURN(-EPERM);
 812
 813         ue = idmap_alloc_entry(rmt_uid, lcl_uid);
 814         if (!ue)
 815                 RETURN(-ENOMEM);
 816         ge = idmap_alloc_entry(rmt_gid, lcl_gid);
 817         if (!ge) {
 818                 idmap_free_entry(ue);
 819                 RETURN(-ENOMEM);
 820         }
 821
 822         spin_lock(&tbl->mit_lock);
 823
 824         if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
 825                                tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
 826                                ue, "UID mapping")) {
 827                 idmap_free_entry(ue);
 828         }
 829
 830         if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
 831                                tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
 832                                ge, "GID mapping")) {
 833                 idmap_free_entry(ge);
 834         }
 835
 836         spin_unlock(&tbl->mit_lock);
 837         RETURN(0);
 838 }
 839
 840 int mds_idmap_del(struct mds_idmap_table *tbl,
 841                   uid_t rmt_uid, uid_t lcl_uid,
 842                   gid_t rmt_gid, gid_t lcl_gid)
 843 {
 844         ENTRY;
 845
 846         if (!tbl)
 847                 RETURN(0);
 848
 849         spin_lock(&tbl->mit_lock);
 850         idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
 851                            tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
 852                            rmt_uid, lcl_uid);
 853         idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
 854                            tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
 855                            rmt_gid, lcl_gid);
 856         spin_unlock(&tbl->mit_lock);
 857         RETURN(0);
 858 }
 859
 860 static
 861 __u32 idmap_lookup_id(struct list_head *hash, int reverse, __u32 id)
 862 {
 863         struct list_head *head = &hash[MDS_IDMAP_HASHFUNC(id)];
 864         struct mds_idmap_entry *e;
 865
 866         if (!reverse) {
 867                 list_for_each_entry(e, head, rmt_hash) {
 868                         if (e->rmt_id == id)
 869                                 return e->lcl_id;
 870                 }
 871                 return MDS_IDMAP_NOTFOUND;
 872         } else {
 873                 list_for_each_entry(e, head, lcl_hash) {
 874                         if (e->lcl_id == id)
 875                                 return e->rmt_id;
 876                 }
 877                 return MDS_IDMAP_NOTFOUND;
 878         }
 879 }
 880
 881 int mds_idmap_lookup_uid(struct mds_idmap_table *tbl, int reverse, uid_t uid)
 882 {
 883         struct list_head *hash;
 884
 885         if (!tbl)
 886                 return MDS_IDMAP_NOTFOUND;
 887
 888         if (!reverse)
 889                 hash = tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX];
 890         else
 891                 hash = tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX];
 892
 893         spin_lock(&tbl->mit_lock);
 894         uid = idmap_lookup_id(hash, reverse, uid);
 895         spin_unlock(&tbl->mit_lock);
 896
 897         return uid;
 898 }
 899
 900 int mds_idmap_lookup_gid(struct mds_idmap_table *tbl, int reverse, gid_t gid)
 901 {
 902         struct list_head *hash;
 903
 904         if (!tbl)
 905                 return MDS_IDMAP_NOTFOUND;
 906
 907         if (!reverse)
 908                 hash = tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX];
 909         else
 910                 hash = tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX];
 911
 912         spin_lock(&tbl->mit_lock);
 913         gid = idmap_lookup_id(hash, reverse, gid);
 914         spin_unlock(&tbl->mit_lock);
 915
 916         return gid;
 917 }
 918
 919 struct mds_idmap_table *mds_idmap_alloc()
 920 {
 921         struct mds_idmap_table *tbl;
 922         int i, j;
 923
 924         OBD_ALLOC(tbl, sizeof(*tbl));
 925         if (!tbl)
 926                 return NULL;
 927
 928         spin_lock_init(&tbl->mit_lock);
 929         for (i = 0; i < MDS_IDMAP_N_HASHES; i++)
 930                 for (j = 0; j < MDS_IDMAP_HASHSIZE; j++)
 931                         INIT_LIST_HEAD(&tbl->mit_idmaps[i][j]);
 932
 933         return tbl;
 934 }
 935
 936 static void idmap_clear_rmt_hash(struct list_head *list)
 937 {
 938         struct mds_idmap_entry *e;
 939         int i;
 940
 941         for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
 942                 while (!list_empty(&list[i])) {
 943                         e = list_entry(list[i].next, struct mds_idmap_entry,
 944                                        rmt_hash);
 945                         idmap_free_entry(e);
 946                 }
 947         }
 948 }
 949
 950 void mds_idmap_free(struct mds_idmap_table *tbl)
 951 {
 952         int i;
 953
 954         spin_lock(&tbl->mit_lock);
 955         idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX]);
 956         idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX]);
 957
 958         /* paranoid checking */
 959         for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
 960                 LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX][i]));
 961                 LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX][i]));
 962         }
 963         spin_unlock(&tbl->mit_lock);
 964
 965         OBD_FREE(tbl, sizeof(*tbl));
 966 }
 967
 968 /*********************************
 969  * helpers doing mapping for MDS *
 970  *********************************/
 971
 972 /*
 973  * we allow remote setuid/setgid to an "authencated" one,
 974  * this policy probably change later.
 975  */
 976 static
 977 int mds_req_secdesc_do_map(struct mds_export_data *med,
 978                            struct mds_req_sec_desc *rsd)
 979 {
 980         struct mds_idmap_table *idmap = med->med_idmap;
 981         uid_t uid, fsuid;
 982         gid_t gid, fsgid;
 983
 984         uid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_uid);
 985         if (uid == MDS_IDMAP_NOTFOUND) {
 986                 CERROR("can't find map for uid %u\n", rsd->rsd_uid);
 987                 return -EPERM;
 988         }
 989
 990         if (rsd->rsd_uid == rsd->rsd_fsuid)
 991                 fsuid = uid;
 992         else {
 993                 fsuid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_fsuid);
 994                 if (fsuid == MDS_IDMAP_NOTFOUND) {
 995                         CERROR("can't find map for fsuid %u\n", rsd->rsd_fsuid);
 996                         return -EPERM;
 997                 }
 998         }
 999
1000         gid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_gid);
1001         if (gid == MDS_IDMAP_NOTFOUND) {
1002                 CERROR("can't find map for gid %u\n", rsd->rsd_gid);
1003                 return -EPERM;
1004         }
1005
1006         if (rsd->rsd_gid == rsd->rsd_fsgid)
1007                 fsgid = gid;
1008         else {
1009                 fsgid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_fsgid);
1010                 if (fsgid == MDS_IDMAP_NOTFOUND) {
1011                         CERROR("can't find map for fsgid %u\n", rsd->rsd_fsgid);
1012                         return -EPERM;
1013                 }
1014         }
1015
1016         rsd->rsd_uid = uid;
1017         rsd->rsd_gid = gid;
1018         rsd->rsd_fsuid = fsuid;
1019         rsd->rsd_fsgid = fsgid;
1020
1021         return 0;
1022 }
1023
1024 void mds_body_do_reverse_map(struct mds_export_data *med,
1025                              struct mds_body *body)
1026 {
1027         uid_t uid;
1028         gid_t gid;
1029
1030         if (!med->med_remote)
1031                 return;
1032
1033         ENTRY;
1034         if (body->valid & OBD_MD_FLUID) {
1035                 uid = mds_idmap_lookup_uid(med->med_idmap, 1, body->uid);
1036                 if (uid == MDS_IDMAP_NOTFOUND) {
1037                         uid = med->med_nllu;
1038                         if (body->valid & OBD_MD_FLMODE) {
1039                                 body->mode = (body->mode & ~S_IRWXU) |
1040                                              ((body->mode & S_IRWXO) << 6);
1041                         }
1042                 }
1043                 body->uid = uid;
1044         }
1045         if (body->valid & OBD_MD_FLGID) {
1046                 gid = mds_idmap_lookup_gid(med->med_idmap, 1, body->gid);
1047                 if (gid == MDS_IDMAP_NOTFOUND) {
1048                         gid = med->med_nllg;
1049                         if (body->valid & OBD_MD_FLMODE) {
1050                                 body->mode = (body->mode & ~S_IRWXG) |
1051                                              ((body->mode & S_IRWXO) << 3);
1052                         }
1053                 }
1054                 body->gid = gid;
1055         }
1056
1057         EXIT;
1058 }
1059
1060 /**********************
1061  * MDS ucred handling *
1062  **********************/
1063
1064 static inline void drop_ucred_ginfo(struct lvfs_ucred *ucred)
1065 {
1066         if (ucred->luc_ginfo) {
1067                 put_group_info(ucred->luc_ginfo);
1068                 ucred->luc_ginfo = NULL;
1069         }
1070 }
1071
1072 static inline void drop_ucred_lsd(struct lvfs_ucred *ucred)
1073 {
1074         if (ucred->luc_lsd) {
1075                 mds_put_lsd(ucred->luc_lsd);
1076                 ucred->luc_lsd = NULL;
1077         }
1078 }
1079
1080 /*
1081  * the heart of the uid/gid handling and security checking.
1082  *
1083  * root could set any group_info if we allowed setgroups, while
1084  * normal user only could 'reduce' their group members -- which
1085  * is somewhat expensive.
1086  *
1087  * authenticated as mds user (using mds service credential) could
1088  * bypass all checkings.
1089  */
1090 int mds_init_ucred(struct lvfs_ucred *ucred,
1091                    struct ptlrpc_request *req,
1092                    struct mds_req_sec_desc *rsd)
1093 {
1094         struct mds_obd *mds = &req->rq_export->exp_obd->u.mds;
1095         struct mds_export_data *med = &req->rq_export->u.eu_mds_data;
1096         struct lustre_sec_desc *lsd;
1097         ptl_nid_t peernid = req->rq_peer.peer_id.nid;
1098         struct group_info *gnew;
1099         unsigned int setuid, setgid, strong_sec, root_squashed;
1100         __u32 lsd_perms;
1101         ENTRY;
1102
1103         LASSERT(ucred);
1104         LASSERT(rsd);
1105         LASSERT(rsd->rsd_ngroups <= LUSTRE_MAX_GROUPS);
1106
1107         if (SEC_FLAVOR_MAJOR(req->rq_req_secflvr) == PTLRPCS_FLVR_MAJOR_GSS &&
1108             (SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_AUTH ||
1109              SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_PRIV))
1110                 strong_sec = 1;
1111         else
1112                 strong_sec = 0;
1113
1114         LASSERT(!(req->rq_remote_realm && !strong_sec));
1115
1116         if (strong_sec && req->rq_auth_uid == -1) {
1117                 CWARN("user not authenticated, deny access\n");
1118                 RETURN(-EPERM);
1119         }
1120
1121         /* sanity check: if we use strong authentication, we expect the
1122          * uid which client claimed is true.
1123          * not apply to special mds user .
1124          */
1125         if (!req->rq_auth_usr_mds && strong_sec) {
1126                 if (!med->med_remote) {
1127                         if (req->rq_auth_uid != rsd->rsd_uid) {
1128                                 CERROR("local client "LPU64": auth uid %u "
1129                                        "while client claim %u:%u/%u:%u\n",
1130                                        peernid, req->rq_auth_uid,
1131                                        rsd->rsd_uid, rsd->rsd_gid,
1132                                        rsd->rsd_fsuid, rsd->rsd_fsgid);
1133                                 RETURN(-EPERM);
1134                         }
1135                 } else {
1136                         if (req->rq_mapped_uid == MDS_IDMAP_NOTFOUND) {
1137                                 CWARN("no mapping found, deny\n");
1138                                 RETURN(-EPERM);
1139                         }
1140
1141                         if (mds_req_secdesc_do_map(med, rsd))
1142                                 RETURN(-EPERM);
1143
1144                         if (req->rq_mapped_uid != rsd->rsd_uid) {
1145                                 CERROR("remote client "LPU64": auth uid %u "
1146                                        "while client claim %u:%u/%u:%u\n",
1147                                        peernid, req->rq_auth_uid,
1148                                        rsd->rsd_uid, rsd->rsd_gid,
1149                                        rsd->rsd_fsuid, rsd->rsd_fsgid);
1150                                 RETURN(-EPERM);
1151                         }
1152                 }
1153         }
1154
1155         /* now LSD come into play */
1156         ucred->luc_ginfo = NULL;
1157         ucred->luc_lsd = lsd = mds_get_lsd(rsd->rsd_uid);
1158
1159         if (!lsd) {
1160                 CERROR("Deny access without LSD: uid %d\n", rsd->rsd_uid);
1161                 RETURN(-EPERM);
1162         }
1163
1164         lsd_perms = mds_lsd_get_perms(lsd, med->med_remote, 0, peernid);
1165
1166         /* check setuid/setgid permissions.
1167          * again not apply to special mds user.
1168          */
1169         if (!req->rq_auth_usr_mds) {
1170                 /* find out the setuid/setgid attempt */
1171                 setuid = (rsd->rsd_uid != rsd->rsd_fsuid);
1172                 setgid = (rsd->rsd_gid != rsd->rsd_fsgid ||
1173                           rsd->rsd_gid != lsd->lsd_gid);
1174
1175                 /* check permission of setuid */
1176                 if (setuid && !(lsd_perms & LSD_PERM_SETUID)) {
1177                         CWARN("mds blocked setuid attempt (%u -> %u) "
1178                               "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_fsuid,
1179                               peernid);
1180                         RETURN(-EPERM);
1181                 }
1182
1183                 /* check permission of setgid */
1184                 if (setgid && !(lsd_perms & LSD_PERM_SETGID)) {
1185                         CWARN("mds blocked setgid attempt (%u:%u/%u:%u -> %u) "
1186                               "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_gid,
1187                               rsd->rsd_fsuid, rsd->rsd_fsgid, lsd->lsd_gid,
1188                               peernid);
1189                         RETURN(-EPERM);
1190                 }
1191         }
1192
1193         root_squashed = mds_squash_root(mds, rsd, &peernid);
1194
1195         /* remove privilege for non-root user */
1196         if (rsd->rsd_fsuid)
1197                 rsd->rsd_cap &= ~CAP_FS_MASK;
1198
1199         /* by now every fields other than groups in rsd have been granted */
1200         ucred->luc_nid = peernid;
1201         ucred->luc_uid = rsd->rsd_uid;
1202         ucred->luc_gid = rsd->rsd_gid;
1203         ucred->luc_fsuid = rsd->rsd_fsuid;
1204         ucred->luc_fsgid = rsd->rsd_fsgid;
1205         ucred->luc_cap = rsd->rsd_cap;
1206
1207         /* don't use any supplementary group if we squashed root.
1208          * XXX The exact behavior of root_squash is not defined, we just
1209          * keep the reminder here */
1210         if (root_squashed)
1211                 RETURN(0);
1212
1213         /* install groups from LSD */
1214         if (lsd->lsd_ginfo) {
1215                 ucred->luc_ginfo = lsd->lsd_ginfo;
1216                 get_group_info(ucred->luc_ginfo);
1217         }
1218
1219         /* everything is done if we don't allow setgroups, or it is
1220          * from remote client (which implies forced to be no-setgroups).
1221          *
1222          * Note: remote user's supplementary groups sent along the request
1223          * (if any) are all ignored, but we make the mapped local user's
1224          * supplementary groups take effect.
1225          */
1226         if (med->med_remote || !(lsd_perms & LSD_PERM_SETGRP))
1227                 RETURN(0);
1228
1229         /* root could set any groups as he want (if allowed), normal
1230          * users only could reduce his group array.
1231          */
1232         if (ucred->luc_uid == 0) {
1233                 drop_ucred_ginfo(ucred);
1234
1235                 if (rsd->rsd_ngroups == 0)
1236                         RETURN(0);
1237
1238                 gnew = groups_alloc(rsd->rsd_ngroups);
1239                 if (!gnew) {
1240                         CERROR("out of memory\n");
1241                         drop_ucred_lsd(ucred);
1242                         RETURN(-ENOMEM);
1243                 }
1244                 groups_from_buffer(gnew, rsd->rsd_groups);
1245                 groups_sort(gnew); /* don't rely on client doing this */
1246
1247                 ucred->luc_ginfo = gnew;
1248         } else {
1249                 __u32 set = 0, cur = 0;
1250                 struct group_info *ginfo = ucred->luc_ginfo;
1251
1252                 if (!ginfo)
1253                         RETURN(0);
1254
1255                 /* Note: freeing a group_info count on 'nblocks' instead of
1256                  * 'ngroups', thus we can safely alloc enough buffer and reduce
1257                  * and ngroups number later.
1258                  */
1259                 gnew = groups_alloc(rsd->rsd_ngroups);
1260                 if (!gnew) {
1261                         CERROR("out of memory\n");
1262                         drop_ucred_ginfo(ucred);
1263                         drop_ucred_lsd(ucred);
1264                         RETURN(-ENOMEM);
1265                 }
1266
1267                 while (cur < rsd->rsd_ngroups) {
1268                         if (groups_search(ginfo, rsd->rsd_groups[cur])) {
1269                                 GROUP_AT(gnew, set) = rsd->rsd_groups[cur];
1270                                 set++;
1271                         }
1272                         cur++;
1273                 }
1274                 gnew->ngroups = set;
1275
1276                 put_group_info(ucred->luc_ginfo);
1277                 ucred->luc_ginfo = gnew;
1278         }
1279         RETURN(0);
1280 }
1281
1282 void mds_exit_ucred(struct lvfs_ucred *ucred)
1283 {
1284         ENTRY;
1285         drop_ucred_ginfo(ucred);
1286         drop_ucred_lsd(ucred);
1287         EXIT;
1288 }