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 crypto_key_md *md_key;
 243         struct crypto_key *ckey;
 244         __u32 buflen, *sizep;
 245         void *buf;
 246         int size, rc = 0;
 247         ENTRY;
 248
 249         sizep = lustre_msg_buf(repmsg, (*offset)++, 4);
 250         if (!sizep) {
 251                 CERROR("can't locate returned ckey size buf\n");
 252                 RETURN(-EPROTO);
 253         }
 254         *sizep = cpu_to_le32(sizeof(*ckey));
 255
 256         OBD_ALLOC(md_key, sizeof(*md_key));
 257
 258         buflen = repmsg->buflens[*offset];
 259         buf = lustre_msg_buf(repmsg, (*offset)++, buflen);
 260
 261         size = fsfilt_get_md(obd, inode, md_key, sizeof(*md_key),
 262                            EA_KEY);
 263         if (size <= 0) {
 264                 if (size < 0)
 265                         CERROR("Can not get gskey from MDS ino %lu rc %d\n",
 266                                 inode->i_ino, size);
 267                 GOTO(out, rc = size);
 268         }
 269         if (le32_to_cpu(md_key->md_magic) != MD_KEY_MAGIC) {
 270                 CDEBUG(D_INFO, "given match %x != magic %x\n",
 271                        md_key->md_magic, MD_KEY_MAGIC);
 272                 GOTO(out, rc = 0);
 273         }
 274
 275         CDEBUG(D_INFO, "get key %s mac %s for ino %lu  size %d \n",
 276                md_key->md_ck.ck_key, md_key->md_ck.ck_mac, inode->i_ino, size);
 277         ckey=(struct crypto_key*)buf;
 278
 279         memcpy(ckey, &md_key->md_ck, sizeof(*ckey));
 280         body->valid |= OBD_MD_FLKEY;
 281 out:
 282         OBD_FREE(md_key, sizeof(*md_key));
 283         RETURN(rc);
 284 }
 285
 286 static int mds_get_gskey(struct inode *inode, struct crypto_key *ckey)
 287 {
 288         LASSERT(ckey);
 289         /*tmp create gs key here*/
 290         LASSERT(ckey->ck_type == MKS_TYPE);
 291         get_random_bytes(ckey->ck_key, KEY_SIZE);
 292         RETURN(0);
 293 }
 294
 295 int mds_set_gskey(struct obd_device *obd, void *handle,
 296                   struct inode *inode, void *key, int key_len,
 297                   int valid)
 298 {
 299         struct crypto_key_md *md_key = NULL;
 300         struct crypto_key *ckey = (struct crypto_key *)key;
 301         int rc = 0;
 302         ENTRY;
 303
 304         if (!ckey)
 305                 RETURN(0);
 306
 307         LASSERT(ckey->ck_type == MKS_TYPE || ckey->ck_type == GKS_TYPE);
 308
 309         OBD_ALLOC(md_key, sizeof(*md_key));
 310         if (ckey->ck_type == MKS_TYPE) {
 311                 mds_get_gskey(inode, ckey);
 312         }
 313
 314         rc = fsfilt_get_md(obd, inode, md_key, sizeof(*md_key),
 315                            EA_KEY);
 316         if (rc < 0)
 317                 GOTO(free, rc);
 318         LASSERT(le32_to_cpu(md_key->md_magic) == MD_KEY_MAGIC ||
 319                 md_key->md_magic == 0);
 320
 321         if (le32_to_cpu(md_key->md_magic) == MD_KEY_MAGIC) {
 322                 CDEBUG(D_INFO, "reset key %s mac %s", md_key->md_ck.ck_mac,
 323                        md_key->md_ck.ck_key);
 324         }
 325
 326         md_key->md_magic = cpu_to_le32(MD_KEY_MAGIC);
 327                 /*get key and mac from request buffer*/
 328         if (valid & ATTR_MAC) {
 329                 memcpy(md_key->md_ck.ck_mac, ckey->ck_mac, MAC_SIZE);
 330                         CDEBUG(D_INFO, "set mac %s for ino %lu \n",
 331                                         md_key->md_ck.ck_mac, inode->i_ino);
 332         }
 333         if (valid & ATTR_KEY) {
 334                 memcpy(md_key->md_ck.ck_key, ckey->ck_key, KEY_SIZE);
 335                        CDEBUG(D_INFO, "set key %s for ino %lu \n",
 336                                         md_key->md_ck.ck_key, inode->i_ino);
 337         }
 338         rc = fsfilt_set_md(obd, inode, handle, md_key, sizeof(*md_key), EA_KEY);
 339 free:
 340         if (md_key)
 341                 OBD_FREE(md_key, sizeof(*md_key));
 342         RETURN(rc);
 343 }
 344
 345 int mds_set_crypto_type(struct obd_device *obd, void *val, __u32 vallen)
 346 {
 347         struct mds_obd *mds = &obd->u.mds;
 348         ENTRY;
 349         if (vallen >= strlen("mks") &&
 350              memcmp(val, "mks", vallen) == 0) {
 351                 mds->mds_crypto_type = MKS_TYPE;
 352                 CDEBUG(D_IOCTL, "mks type\n");
 353         }
 354         if (vallen >= strlen("gks") &&
 355              memcmp(val, "gks", vallen) == 0) {
 356                 mds->mds_crypto_type = GKS_TYPE;
 357                 CDEBUG(D_IOCTL, "gks type \n");
 358         }
 359         RETURN(0);
 360 }
 361
 362 /* Note that we can copy all of the fields, just some will not be "valid" */
 363 void mds_pack_inode2body(struct obd_device *obd, struct mds_body *b,
 364                          struct inode *inode, int fid)
 365 {
 366         b->valid |= OBD_MD_FLID | OBD_MD_FLCTIME | OBD_MD_FLUID |
 367                 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLTYPE |
 368                 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLGENER |
 369                 OBD_MD_FLATIME | OBD_MD_FLMTIME; /* bug 2020 */
 370
 371         if (!S_ISREG(inode->i_mode)) {
 372                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
 373                         OBD_MD_FLATIME | OBD_MD_FLMTIME |
 374                         OBD_MD_FLRDEV;
 375         }
 376         b->atime = LTIME_S(inode->i_atime);
 377         b->mtime = LTIME_S(inode->i_mtime);
 378         b->ctime = LTIME_S(inode->i_ctime);
 379         b->mode = inode->i_mode;
 380         b->size = inode->i_size;
 381         b->blocks = inode->i_blocks;
 382         b->uid = inode->i_uid;
 383         b->gid = inode->i_gid;
 384         b->flags = inode->i_flags;
 385         b->rdev = inode->i_rdev;
 386
 387         /* Return the correct link count for orphan inodes */
 388         if (mds_inode_is_orphan(inode)) {
 389                 b->nlink = 0;
 390         } else if (S_ISDIR(inode->i_mode)) {
 391                 b->nlink = 1;
 392         } else {
 393                 b->nlink = inode->i_nlink;
 394         }
 395
 396         if (fid)
 397                 b->valid |= OBD_MD_FID;
 398
 399         mds_pack_inode2id(obd, &b->id1, inode, fid);
 400 }
 401
 402 /* unpacking */
 403 static int mds_setattr_unpack(struct ptlrpc_request *req, int offset,
 404                               struct mds_update_record *r)
 405 {
 406         struct iattr *attr = &r->ur_iattr;
 407         struct mds_rec_setattr *rec;
 408         ENTRY;
 409
 410         rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
 411                                  lustre_swab_mds_rec_setattr);
 412         if (rec == NULL)
 413                 RETURN (-EFAULT);
 414
 415         r->ur_id1 = &rec->sa_id;
 416         r->ur_flags = rec->sa_flags;
 417         attr->ia_valid = rec->sa_valid;
 418         attr->ia_mode = rec->sa_mode;
 419         attr->ia_uid = rec->sa_uid;
 420         attr->ia_gid = rec->sa_gid;
 421         attr->ia_size = rec->sa_size;
 422         LTIME_S(attr->ia_atime) = rec->sa_atime;
 423         LTIME_S(attr->ia_mtime) = rec->sa_mtime;
 424         LTIME_S(attr->ia_ctime) = rec->sa_ctime;
 425         attr->ia_attr_flags = rec->sa_attr_flags;
 426
 427         LASSERT_REQSWAB(req, offset + 1);
 428         if (req->rq_reqmsg->bufcount > offset + 1) {
 429                 r->ur_eadata = lustre_msg_buf(req->rq_reqmsg,
 430                                               offset + 1, 0);
 431                 if (r->ur_eadata == NULL)
 432                         RETURN (-EFAULT);
 433                 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 1];
 434         }
 435
 436         if (req->rq_reqmsg->bufcount > offset + 2) {
 437                 r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
 438                 if (r->ur_ea2data == NULL)
 439                         RETURN (-EFAULT);
 440
 441                 r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 2];
 442         }
 443
 444         if (req->rq_reqmsg->bufcount > offset + 3) {
 445                 r->ur_ea3data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
 446                 if (r->ur_ea3data == NULL)
 447                         RETURN (-EFAULT);
 448
 449                 r->ur_ea3datalen = req->rq_reqmsg->buflens[offset + 3];
 450         }
 451
 452         RETURN(0);
 453 }
 454
 455 static int mds_create_unpack(struct ptlrpc_request *req, int offset,
 456                              struct mds_update_record *r)
 457 {
 458         struct mds_rec_create *rec;
 459         ENTRY;
 460
 461         rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
 462                                  lustre_swab_mds_rec_create);
 463         if (rec == NULL)
 464                 RETURN (-EFAULT);
 465
 466         r->ur_id1 = &rec->cr_id;
 467         r->ur_id2 = &rec->cr_replayid;
 468         r->ur_mode = rec->cr_mode;
 469         r->ur_rdev = rec->cr_rdev;
 470         r->ur_time = rec->cr_time;
 471         r->ur_flags = rec->cr_flags;
 472
 473         LASSERT_REQSWAB(req, offset + 1);
 474         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 475         if (r->ur_name == NULL)
 476                 RETURN(-EFAULT);
 477         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 478
 479         LASSERT_REQSWAB(req, offset + 2);
 480         if (req->rq_reqmsg->bufcount > offset + 2) {
 481                 if (S_ISLNK(r->ur_mode)) {
 482                         r->ur_tgt = lustre_msg_string(req->rq_reqmsg,
 483                                                       offset + 2, 0);
 484                         if (r->ur_tgt == NULL)
 485                                 RETURN(-EFAULT);
 486                         r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
 487                 } else if (S_ISDIR(r->ur_mode) ) {
 488                         /* Stripe info for mkdir - just a 16bit integer */
 489                         if (req->rq_reqmsg->buflens[offset + 2] != 2) {
 490                                 CERROR("mkdir stripe info does not match "
 491                                        "expected size %d vs 2\n",
 492                                        req->rq_reqmsg->buflens[offset + 2]);
 493                                 RETURN(-EINVAL);
 494                         }
 495                         r->ur_eadata = lustre_swab_buf(req->rq_reqmsg,
 496                                                        offset + 2, 2,
 497                                                        __swab16s);
 498                         r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
 499                 } else if (S_ISREG(r->ur_mode)){
 500                         r->ur_eadata = lustre_msg_buf(req->rq_reqmsg,
 501                                                       offset + 2, 0);
 502                         r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
 503                 } else {
 504                         /* Hm, no other users so far? */
 505                         LBUG();
 506                 }
 507         }
 508         RETURN(0);
 509 }
 510
 511 static int mds_link_unpack(struct ptlrpc_request *req, int offset,
 512                            struct mds_update_record *r)
 513 {
 514         struct mds_rec_link *rec;
 515         ENTRY;
 516
 517         rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
 518                                  lustre_swab_mds_rec_link);
 519         if (rec == NULL)
 520                 RETURN(-EFAULT);
 521
 522         r->ur_id1 = &rec->lk_id1;
 523         r->ur_id2 = &rec->lk_id2;
 524         r->ur_time = rec->lk_time;
 525         r->ur_flags = rec->lk_flags;
 526
 527         LASSERT_REQSWAB(req, offset + 1);
 528         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 529         if (r->ur_name == NULL)
 530                 RETURN(-EFAULT);
 531         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 532         RETURN(0);
 533 }
 534
 535 static int mds_unlink_unpack(struct ptlrpc_request *req, int offset,
 536                              struct mds_update_record *r)
 537 {
 538         struct mds_rec_unlink *rec;
 539         ENTRY;
 540
 541         rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
 542                                  lustre_swab_mds_rec_unlink);
 543         if (rec == NULL)
 544                 RETURN(-EFAULT);
 545
 546         r->ur_mode = rec->ul_mode;
 547         r->ur_id1 = &rec->ul_id1;
 548         r->ur_id2 = &rec->ul_id2;
 549         r->ur_time = rec->ul_time;
 550         r->ur_flags = rec->ul_flags;
 551
 552         LASSERT_REQSWAB(req, offset + 1);
 553         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 554         if (r->ur_name == NULL)
 555                 RETURN(-EFAULT);
 556         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 557         RETURN(0);
 558 }
 559
 560 static int mds_rename_unpack(struct ptlrpc_request *req, int offset,
 561                              struct mds_update_record *r)
 562 {
 563         struct mds_rec_rename *rec;
 564         ENTRY;
 565
 566         rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
 567                                  lustre_swab_mds_rec_rename);
 568         if (rec == NULL)
 569                 RETURN(-EFAULT);
 570
 571         r->ur_id1 = &rec->rn_id1;
 572         r->ur_id2 = &rec->rn_id2;
 573         r->ur_time = rec->rn_time;
 574         r->ur_flags = rec->rn_flags;
 575
 576         LASSERT_REQSWAB(req, offset + 1);
 577         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 578         if (r->ur_name == NULL)
 579                 RETURN(-EFAULT);
 580         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 581
 582         LASSERT_REQSWAB(req, offset + 2);
 583         r->ur_tgt = lustre_msg_string(req->rq_reqmsg, offset + 2, 0);
 584         if (r->ur_tgt == NULL)
 585                 RETURN(-EFAULT);
 586         r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
 587         RETURN(0);
 588 }
 589
 590 static int mds_open_unpack(struct ptlrpc_request *req, int offset,
 591                            struct mds_update_record *r)
 592 {
 593         struct mds_rec_create *rec;
 594         ENTRY;
 595
 596         rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
 597                                  lustre_swab_mds_rec_create);
 598         if (rec == NULL)
 599                 RETURN(-EFAULT);
 600
 601         r->ur_id1 = &rec->cr_id;
 602         r->ur_id2 = &rec->cr_replayid;
 603         r->ur_mode = rec->cr_mode;
 604         r->ur_rdev = rec->cr_rdev;
 605         r->ur_time = rec->cr_time;
 606         r->ur_flags = rec->cr_flags;
 607
 608         LASSERT_REQSWAB(req, offset + 1);
 609         r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
 610
 611         if (r->ur_name == NULL)
 612                 RETURN(-EFAULT);
 613         r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
 614
 615         LASSERT_REQSWAB(req, offset + 2);
 616         if (req->rq_reqmsg->bufcount > offset + 2) {
 617                 r->ur_eadata = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
 618                 if (r->ur_eadata == NULL)
 619                         RETURN(-EFAULT);
 620                 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
 621         }
 622
 623         if (rec->cr_flags & MDS_OPEN_HAS_KEY) {
 624                 LASSERT(req->rq_reqmsg->bufcount > offset + 3);
 625                 r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
 626                 r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 3];
 627         }
 628         RETURN(0);
 629 }
 630
 631 typedef int (*update_unpacker)(struct ptlrpc_request *req, int offset,
 632                                struct mds_update_record *r);
 633
 634 static update_unpacker mds_unpackers[REINT_MAX + 1] = {
 635         [REINT_SETATTR] mds_setattr_unpack,
 636         [REINT_CREATE] mds_create_unpack,
 637         [REINT_LINK] mds_link_unpack,
 638         [REINT_UNLINK] mds_unlink_unpack,
 639         [REINT_RENAME] mds_rename_unpack,
 640         [REINT_OPEN] mds_open_unpack,
 641 };
 642
 643 int mds_update_unpack(struct ptlrpc_request *req, int offset,
 644                       struct mds_update_record *rec)
 645 {
 646         __u32 *opcodep;
 647         __u32  opcode;
 648         int rc;
 649         ENTRY;
 650
 651         /* NB don't lustre_swab_reqbuf() here. We're just taking a peek and we
 652          * want to leave it to the specific unpacker once we've identified the
 653          * message type. */
 654         opcodep = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*opcodep));
 655         if (opcodep == NULL)
 656                 RETURN(-EFAULT);
 657
 658         opcode = *opcodep;
 659         if (lustre_msg_swabbed(req->rq_reqmsg))
 660                 __swab32s(&opcode);
 661
 662         if (opcode > REINT_MAX ||
 663             mds_unpackers[opcode] == NULL) {
 664                 CERROR("Unexpected opcode %d\n", opcode);
 665                 RETURN(-EFAULT);
 666         }
 667
 668         rec->ur_id1 = NULL;
 669         rec->ur_id2 = NULL;
 670         rec->ur_opcode = opcode;
 671
 672         rc = mds_unpackers[opcode](req, offset, rec);
 673
 674 #if CRAY_PORTALS
 675         rec->ur_fsuid = req->rq_uid;
 676 #endif
 677         RETURN(rc);
 678 }
 679
 680 /*
 681  * here we take simple rule: once uid/fsuid is root, we also squash
 682  * the gid/fsgid, don't care setuid/setgid attributes.
 683  */
 684 static
 685 int mds_squash_root(struct mds_obd *mds, struct mds_req_sec_desc *rsd,
 686                     ptl_nid_t *peernid)
 687 {
 688         if (!mds->mds_squash_uid || *peernid == mds->mds_nosquash_nid)
 689                 return 0;
 690
 691         if (rsd->rsd_uid && rsd->rsd_fsuid)
 692                 return 0;
 693
 694         CDEBUG(D_SEC, "squash req from "LPX64":"
 695                "(%u:%u-%u:%u/%x)=>(%u:%u-%u:%u/%x)\n", *peernid,
 696                 rsd->rsd_uid, rsd->rsd_gid,
 697                 rsd->rsd_fsuid, rsd->rsd_fsgid, rsd->rsd_cap,
 698                 rsd->rsd_uid ? rsd->rsd_uid : mds->mds_squash_uid,
 699                 rsd->rsd_uid ? rsd->rsd_gid : mds->mds_squash_gid,
 700                 rsd->rsd_fsuid ? rsd->rsd_fsuid : mds->mds_squash_uid,
 701                 rsd->rsd_fsuid ? rsd->rsd_fsgid : mds->mds_squash_gid,
 702                 rsd->rsd_cap & ~CAP_FS_MASK);
 703
 704         if (rsd->rsd_uid == 0) {
 705                 rsd->rsd_uid = mds->mds_squash_uid;
 706                 rsd->rsd_gid = mds->mds_squash_gid;
 707         }
 708         if (rsd->rsd_fsuid == 0) {
 709                 rsd->rsd_fsuid = mds->mds_squash_uid;
 710                 rsd->rsd_fsgid = mds->mds_squash_gid;
 711         }
 712         rsd->rsd_cap &= ~CAP_FS_MASK;
 713
 714         return 1;
 715 }
 716
 717 /********************************
 718  * MDS uid/gid mapping handling *
 719  ********************************/
 720
 721 static
 722 struct mds_idmap_entry* idmap_alloc_entry(__u32 rmt_id, __u32 lcl_id)
 723 {
 724         struct mds_idmap_entry *e;
 725
 726         OBD_ALLOC(e, sizeof(*e));
 727         if (!e)
 728                 return NULL;
 729
 730         INIT_LIST_HEAD(&e->rmt_hash);
 731         INIT_LIST_HEAD(&e->lcl_hash);
 732         atomic_set(&e->refcount, 1);
 733         e->rmt_id = rmt_id;
 734         e->lcl_id = lcl_id;
 735
 736         return e;
 737 }
 738
 739 void idmap_free_entry(struct mds_idmap_entry *e)
 740 {
 741         if (!list_empty(&e->rmt_hash))
 742                 list_del(&e->rmt_hash);
 743         if (!list_empty(&e->lcl_hash))
 744                 list_del(&e->lcl_hash);
 745         OBD_FREE(e, sizeof(*e));
 746 }
 747
 748 static
 749 int idmap_insert_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
 750                        struct mds_idmap_entry *new, const char *warn_msg)
 751 {
 752         struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(new->rmt_id)];
 753         struct list_head *lcl_head = &lcl_hash[MDS_IDMAP_HASHFUNC(new->lcl_id)];
 754         struct mds_idmap_entry *e;
 755
 756         list_for_each_entry(e, rmt_head, rmt_hash) {
 757                 if (e->rmt_id == new->rmt_id &&
 758                     e->lcl_id == new->lcl_id) {
 759                         atomic_inc(&e->refcount);
 760                         return 1;
 761                 }
 762                 if (e->rmt_id == new->rmt_id && warn_msg)
 763                         CWARN("%s: rmt id %u already map to %u (new %u)\n",
 764                               warn_msg, e->rmt_id, e->lcl_id, new->lcl_id);
 765                 if (e->lcl_id == new->lcl_id && warn_msg)
 766                         CWARN("%s: lcl id %u already be mapped from %u "
 767                               "(new %u)\n", warn_msg,
 768                               e->lcl_id, e->rmt_id, new->rmt_id);
 769         }
 770
 771         list_add_tail(rmt_head, &new->rmt_hash);
 772         list_add_tail(lcl_head, &new->lcl_hash);
 773         return 0;
 774 }
 775
 776 static
 777 int idmap_remove_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
 778                        __u32 rmt_id, __u32 lcl_id)
 779 {
 780         struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(rmt_id)];
 781         struct mds_idmap_entry *e;
 782
 783         list_for_each_entry(e, rmt_head, rmt_hash) {
 784                 if (e->rmt_id == rmt_id && e->lcl_id == lcl_id) {
 785                         if (atomic_dec_and_test(&e->refcount)) {
 786                                 list_del(&e->rmt_hash);
 787                                 list_del(&e->lcl_hash);
 788                                 OBD_FREE(e, sizeof(*e));
 789                                 return 0;
 790                         } else
 791                                 return 1;
 792                 }
 793         }
 794         return -ENOENT;
 795 }
 796
 797 int mds_idmap_add(struct mds_idmap_table *tbl,
 798                   uid_t rmt_uid, uid_t lcl_uid,
 799                   gid_t rmt_gid, gid_t lcl_gid)
 800 {
 801         struct mds_idmap_entry *ue, *ge;
 802         ENTRY;
 803
 804         if (!tbl)
 805                 RETURN(-EPERM);
 806
 807         ue = idmap_alloc_entry(rmt_uid, lcl_uid);
 808         if (!ue)
 809                 RETURN(-ENOMEM);
 810         ge = idmap_alloc_entry(rmt_gid, lcl_gid);
 811         if (!ge) {
 812                 idmap_free_entry(ue);
 813                 RETURN(-ENOMEM);
 814         }
 815
 816         spin_lock(&tbl->mit_lock);
 817
 818         if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
 819                                tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
 820                                ue, "UID mapping")) {
 821                 idmap_free_entry(ue);
 822         }
 823
 824         if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
 825                                tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
 826                                ge, "GID mapping")) {
 827                 idmap_free_entry(ge);
 828         }
 829
 830         spin_unlock(&tbl->mit_lock);
 831         RETURN(0);
 832 }
 833
 834 int mds_idmap_del(struct mds_idmap_table *tbl,
 835                   uid_t rmt_uid, uid_t lcl_uid,
 836                   gid_t rmt_gid, gid_t lcl_gid)
 837 {
 838         ENTRY;
 839
 840         if (!tbl)
 841                 RETURN(0);
 842
 843         spin_lock(&tbl->mit_lock);
 844         idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
 845                            tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
 846                            rmt_uid, lcl_uid);
 847         idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
 848                            tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
 849                            rmt_gid, lcl_gid);
 850         spin_unlock(&tbl->mit_lock);
 851         RETURN(0);
 852 }
 853
 854 static
 855 __u32 idmap_lookup_id(struct list_head *hash, int reverse, __u32 id)
 856 {
 857         struct list_head *head = &hash[MDS_IDMAP_HASHFUNC(id)];
 858         struct mds_idmap_entry *e;
 859
 860         if (!reverse) {
 861                 list_for_each_entry(e, head, rmt_hash) {
 862                         if (e->rmt_id == id)
 863                                 return e->lcl_id;
 864                 }
 865                 return MDS_IDMAP_NOTFOUND;
 866         } else {
 867                 list_for_each_entry(e, head, lcl_hash) {
 868                         if (e->lcl_id == id)
 869                                 return e->rmt_id;
 870                 }
 871                 return MDS_IDMAP_NOTFOUND;
 872         }
 873 }
 874
 875 int mds_idmap_lookup_uid(struct mds_idmap_table *tbl, int reverse, uid_t uid)
 876 {
 877         struct list_head *hash;
 878
 879         if (!tbl)
 880                 return MDS_IDMAP_NOTFOUND;
 881
 882         if (!reverse)
 883                 hash = tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX];
 884         else
 885                 hash = tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX];
 886
 887         spin_lock(&tbl->mit_lock);
 888         uid = idmap_lookup_id(hash, reverse, uid);
 889         spin_unlock(&tbl->mit_lock);
 890
 891         return uid;
 892 }
 893
 894 int mds_idmap_lookup_gid(struct mds_idmap_table *tbl, int reverse, gid_t gid)
 895 {
 896         struct list_head *hash;
 897
 898         if (!tbl)
 899                 return MDS_IDMAP_NOTFOUND;
 900
 901         if (!reverse)
 902                 hash = tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX];
 903         else
 904                 hash = tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX];
 905
 906         spin_lock(&tbl->mit_lock);
 907         gid = idmap_lookup_id(hash, reverse, gid);
 908         spin_unlock(&tbl->mit_lock);
 909
 910         return gid;
 911 }
 912
 913 struct mds_idmap_table *mds_idmap_alloc()
 914 {
 915         struct mds_idmap_table *tbl;
 916         int i, j;
 917
 918         OBD_ALLOC(tbl, sizeof(*tbl));
 919         if (!tbl)
 920                 return NULL;
 921
 922         spin_lock_init(&tbl->mit_lock);
 923         for (i = 0; i < MDS_IDMAP_N_HASHES; i++)
 924                 for (j = 0; j < MDS_IDMAP_HASHSIZE; j++)
 925                         INIT_LIST_HEAD(&tbl->mit_idmaps[i][j]);
 926
 927         return tbl;
 928 }
 929
 930 static void idmap_clear_rmt_hash(struct list_head *list)
 931 {
 932         struct mds_idmap_entry *e;
 933         int i;
 934
 935         for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
 936                 while (!list_empty(&list[i])) {
 937                         e = list_entry(list[i].next, struct mds_idmap_entry,
 938                                        rmt_hash);
 939                         idmap_free_entry(e);
 940                 }
 941         }
 942 }
 943
 944 void mds_idmap_free(struct mds_idmap_table *tbl)
 945 {
 946         int i;
 947
 948         spin_lock(&tbl->mit_lock);
 949         idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX]);
 950         idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX]);
 951
 952         /* paranoid checking */
 953         for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
 954                 LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX][i]));
 955                 LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX][i]));
 956         }
 957         spin_unlock(&tbl->mit_lock);
 958
 959         OBD_FREE(tbl, sizeof(*tbl));
 960 }
 961
 962 /*********************************
 963  * helpers doing mapping for MDS *
 964  *********************************/
 965
 966 /*
 967  * we allow remote setuid/setgid to an "authencated" one,
 968  * this policy probably change later.
 969  */
 970 static
 971 int mds_req_secdesc_do_map(struct mds_export_data *med,
 972                            struct mds_req_sec_desc *rsd)
 973 {
 974         struct mds_idmap_table *idmap = med->med_idmap;
 975         uid_t uid, fsuid;
 976         gid_t gid, fsgid;
 977
 978         uid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_uid);
 979         if (uid == MDS_IDMAP_NOTFOUND) {
 980                 CERROR("can't find map for uid %u\n", rsd->rsd_uid);
 981                 return -EPERM;
 982         }
 983
 984         if (rsd->rsd_uid == rsd->rsd_fsuid)
 985                 fsuid = uid;
 986         else {
 987                 fsuid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_fsuid);
 988                 if (fsuid == MDS_IDMAP_NOTFOUND) {
 989                         CERROR("can't find map for fsuid %u\n", rsd->rsd_fsuid);
 990                         return -EPERM;
 991                 }
 992         }
 993
 994         gid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_gid);
 995         if (gid == MDS_IDMAP_NOTFOUND) {
 996                 CERROR("can't find map for gid %u\n", rsd->rsd_gid);
 997                 return -EPERM;
 998         }
 999
1000         if (rsd->rsd_gid == rsd->rsd_fsgid)
1001                 fsgid = gid;
1002         else {
1003                 fsgid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_fsgid);
1004                 if (fsgid == MDS_IDMAP_NOTFOUND) {
1005                         CERROR("can't find map for fsgid %u\n", rsd->rsd_fsgid);
1006                         return -EPERM;
1007                 }
1008         }
1009
1010         rsd->rsd_uid = uid;
1011         rsd->rsd_gid = gid;
1012         rsd->rsd_fsuid = fsuid;
1013         rsd->rsd_fsgid = fsgid;
1014
1015         return 0;
1016 }
1017
1018 void mds_body_do_reverse_map(struct mds_export_data *med,
1019                              struct mds_body *body)
1020 {
1021         uid_t uid;
1022         gid_t gid;
1023
1024         if (!med->med_remote)
1025                 return;
1026
1027         ENTRY;
1028         if (body->valid & OBD_MD_FLUID) {
1029                 uid = mds_idmap_lookup_uid(med->med_idmap, 1, body->uid);
1030                 if (uid == MDS_IDMAP_NOTFOUND) {
1031                         uid = med->med_nllu;
1032                         if (body->valid & OBD_MD_FLMODE) {
1033                                 body->mode = (body->mode & ~S_IRWXU) |
1034                                              ((body->mode & S_IRWXO) << 6);
1035                         }
1036                 }
1037                 body->uid = uid;
1038         }
1039         if (body->valid & OBD_MD_FLGID) {
1040                 gid = mds_idmap_lookup_gid(med->med_idmap, 1, body->gid);
1041                 if (gid == MDS_IDMAP_NOTFOUND) {
1042                         gid = med->med_nllg;
1043                         if (body->valid & OBD_MD_FLMODE) {
1044                                 body->mode = (body->mode & ~S_IRWXG) |
1045                                              ((body->mode & S_IRWXO) << 3);
1046                         }
1047                 }
1048                 body->gid = gid;
1049         }
1050
1051         EXIT;
1052 }
1053
1054 /**********************
1055  * MDS ucred handling *
1056  **********************/
1057
1058 static inline void drop_ucred_ginfo(struct lvfs_ucred *ucred)
1059 {
1060         if (ucred->luc_ginfo) {
1061                 put_group_info(ucred->luc_ginfo);
1062                 ucred->luc_ginfo = NULL;
1063         }
1064 }
1065
1066 static inline void drop_ucred_lsd(struct lvfs_ucred *ucred)
1067 {
1068         if (ucred->luc_lsd) {
1069                 mds_put_lsd(ucred->luc_lsd);
1070                 ucred->luc_lsd = NULL;
1071         }
1072 }
1073
1074 /*
1075  * the heart of the uid/gid handling and security checking.
1076  *
1077  * root could set any group_info if we allowed setgroups, while
1078  * normal user only could 'reduce' their group members -- which
1079  * is somewhat expensive.
1080  *
1081  * authenticated as mds user (using mds service credential) could
1082  * bypass all checkings.
1083  */
1084 int mds_init_ucred(struct lvfs_ucred *ucred,
1085                    struct ptlrpc_request *req,
1086                    struct mds_req_sec_desc *rsd)
1087 {
1088         struct mds_obd *mds = &req->rq_export->exp_obd->u.mds;
1089         struct mds_export_data *med = &req->rq_export->u.eu_mds_data;
1090         struct lustre_sec_desc *lsd;
1091         ptl_nid_t peernid = req->rq_peer.peer_id.nid;
1092         struct group_info *gnew;
1093         unsigned int setuid, setgid, strong_sec, root_squashed;
1094         __u32 lsd_perms;
1095         ENTRY;
1096
1097         LASSERT(ucred);
1098         LASSERT(rsd);
1099         LASSERT(rsd->rsd_ngroups <= LUSTRE_MAX_GROUPS);
1100
1101         if (SEC_FLAVOR_MAJOR(req->rq_req_secflvr) == PTLRPCS_FLVR_MAJOR_GSS &&
1102             (SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_AUTH ||
1103              SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_PRIV))
1104                 strong_sec = 1;
1105         else
1106                 strong_sec = 0;
1107
1108         LASSERT(!(req->rq_remote_realm && !strong_sec));
1109
1110         if (strong_sec && req->rq_auth_uid == -1) {
1111                 CWARN("user not authenticated, deny access\n");
1112                 RETURN(-EPERM);
1113         }
1114
1115         /* sanity check: if we use strong authentication, we expect the
1116          * uid which client claimed is true.
1117          * not apply to special mds user .
1118          */
1119         if (!req->rq_auth_usr_mds && strong_sec) {
1120                 if (!med->med_remote) {
1121                         if (req->rq_auth_uid != rsd->rsd_uid) {
1122                                 CERROR("local client "LPU64": auth uid %u "
1123                                        "while client claim %u:%u/%u:%u\n",
1124                                        peernid, req->rq_auth_uid,
1125                                        rsd->rsd_uid, rsd->rsd_gid,
1126                                        rsd->rsd_fsuid, rsd->rsd_fsgid);
1127                                 RETURN(-EPERM);
1128                         }
1129                 } else {
1130                         if (req->rq_mapped_uid == MDS_IDMAP_NOTFOUND) {
1131                                 CWARN("no mapping found, deny\n");
1132                                 RETURN(-EPERM);
1133                         }
1134
1135                         if (mds_req_secdesc_do_map(med, rsd))
1136                                 RETURN(-EPERM);
1137
1138                         if (req->rq_mapped_uid != rsd->rsd_uid) {
1139                                 CERROR("remote client "LPU64": auth uid %u "
1140                                        "while client claim %u:%u/%u:%u\n",
1141                                        peernid, req->rq_auth_uid,
1142                                        rsd->rsd_uid, rsd->rsd_gid,
1143                                        rsd->rsd_fsuid, rsd->rsd_fsgid);
1144                                 RETURN(-EPERM);
1145                         }
1146                 }
1147         }
1148
1149         /* now LSD come into play */
1150         ucred->luc_ginfo = NULL;
1151         ucred->luc_lsd = lsd = mds_get_lsd(rsd->rsd_uid);
1152
1153         if (!lsd) {
1154                 CERROR("Deny access without LSD: uid %d\n", rsd->rsd_uid);
1155                 RETURN(-EPERM);
1156         }
1157
1158         lsd_perms = mds_lsd_get_perms(lsd, med->med_remote, 0, peernid);
1159
1160         /* check setuid/setgid permissions.
1161          * again not apply to special mds user.
1162          */
1163         if (!req->rq_auth_usr_mds) {
1164                 /* find out the setuid/setgid attempt */
1165                 setuid = (rsd->rsd_uid != rsd->rsd_fsuid);
1166                 setgid = (rsd->rsd_gid != rsd->rsd_fsgid ||
1167                           rsd->rsd_gid != lsd->lsd_gid);
1168
1169                 /* check permission of setuid */
1170                 if (setuid && !(lsd_perms & LSD_PERM_SETUID)) {
1171                         CWARN("mds blocked setuid attempt (%u -> %u) "
1172                               "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_fsuid,
1173                               peernid);
1174                         RETURN(-EPERM);
1175                 }
1176
1177                 /* check permission of setgid */
1178                 if (setgid && !(lsd_perms & LSD_PERM_SETGID)) {
1179                         CWARN("mds blocked setgid attempt (%u:%u/%u:%u -> %u) "
1180                               "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_gid,
1181                               rsd->rsd_fsuid, rsd->rsd_fsgid, lsd->lsd_gid,
1182                               peernid);
1183                         RETURN(-EPERM);
1184                 }
1185         }
1186
1187         root_squashed = mds_squash_root(mds, rsd, &peernid);
1188
1189         /* remove privilege for non-root user */
1190         if (rsd->rsd_fsuid)
1191                 rsd->rsd_cap &= ~CAP_FS_MASK;
1192
1193         /* by now every fields other than groups in rsd have been granted */
1194         ucred->luc_nid = peernid;
1195         ucred->luc_uid = rsd->rsd_uid;
1196         ucred->luc_gid = rsd->rsd_gid;
1197         ucred->luc_fsuid = rsd->rsd_fsuid;
1198         ucred->luc_fsgid = rsd->rsd_fsgid;
1199         ucred->luc_cap = rsd->rsd_cap;
1200
1201         /* don't use any supplementary group if we squashed root.
1202          * XXX The exact behavior of root_squash is not defined, we just
1203          * keep the reminder here */
1204         if (root_squashed)
1205                 RETURN(0);
1206
1207         /* install groups from LSD */
1208         if (lsd->lsd_ginfo) {
1209                 ucred->luc_ginfo = lsd->lsd_ginfo;
1210                 get_group_info(ucred->luc_ginfo);
1211         }
1212
1213         /* everything is done if we don't allow setgroups, or it is
1214          * from remote client (which implies forced to be no-setgroups).
1215          *
1216          * Note: remote user's supplementary groups sent along the request
1217          * (if any) are all ignored, but we make the mapped local user's
1218          * supplementary groups take effect.
1219          */
1220         if (med->med_remote || !(lsd_perms & LSD_PERM_SETGRP))
1221                 RETURN(0);
1222
1223         /* root could set any groups as he want (if allowed), normal
1224          * users only could reduce his group array.
1225          */
1226         if (ucred->luc_uid == 0) {
1227                 drop_ucred_ginfo(ucred);
1228
1229                 if (rsd->rsd_ngroups == 0)
1230                         RETURN(0);
1231
1232                 gnew = groups_alloc(rsd->rsd_ngroups);
1233                 if (!gnew) {
1234                         CERROR("out of memory\n");
1235                         drop_ucred_lsd(ucred);
1236                         RETURN(-ENOMEM);
1237                 }
1238                 groups_from_buffer(gnew, rsd->rsd_groups);
1239                 groups_sort(gnew); /* don't rely on client doing this */
1240
1241                 ucred->luc_ginfo = gnew;
1242         } else {
1243                 __u32 set = 0, cur = 0;
1244                 struct group_info *ginfo = ucred->luc_ginfo;
1245
1246                 if (!ginfo)
1247                         RETURN(0);
1248
1249                 /* Note: freeing a group_info count on 'nblocks' instead of
1250                  * 'ngroups', thus we can safely alloc enough buffer and reduce
1251                  * and ngroups number later.
1252                  */
1253                 gnew = groups_alloc(rsd->rsd_ngroups);
1254                 if (!gnew) {
1255                         CERROR("out of memory\n");
1256                         drop_ucred_ginfo(ucred);
1257                         drop_ucred_lsd(ucred);
1258                         RETURN(-ENOMEM);
1259                 }
1260
1261                 while (cur < rsd->rsd_ngroups) {
1262                         if (groups_search(ginfo, rsd->rsd_groups[cur])) {
1263                                 GROUP_AT(gnew, set) = rsd->rsd_groups[cur];
1264                                 set++;
1265                         }
1266                         cur++;
1267                 }
1268                 gnew->ngroups = set;
1269
1270                 put_group_info(ucred->luc_ginfo);
1271                 ucred->luc_ginfo = gnew;
1272         }
1273         RETURN(0);
1274 }
1275
1276 void mds_exit_ucred(struct lvfs_ucred *ucred)
1277 {
1278         ENTRY;
1279         drop_ucred_ginfo(ucred);
1280         drop_ucred_lsd(ucred);
1281         EXIT;
1282 }