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