drivers/staging/lustre/lustre/ptlrpc/gss/gss_svc_upcall.c

   1 /*
   2  * Modifications for Lustre
   3  *
   4  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
   5  *
   6  * Copyright (c) 2011, 2012, Intel Corporation.
   7  *
   8  * Author: Eric Mei <ericm@clusterfs.com>
   9  */
  10
  11 /*
  12  * Neil Brown <neilb@cse.unsw.edu.au>
  13  * J. Bruce Fields <bfields@umich.edu>
  14  * Andy Adamson <andros@umich.edu>
  15  * Dug Song <dugsong@monkey.org>
  16  *
  17  * RPCSEC_GSS server authentication.
  18  * This implements RPCSEC_GSS as defined in rfc2203 (rpcsec_gss) and rfc2078
  19  * (gssapi)
  20  *
  21  * The RPCSEC_GSS involves three stages:
  22  *  1/ context creation
  23  *  2/ data exchange
  24  *  3/ context destruction
  25  *
  26  * Context creation is handled largely by upcalls to user-space.
  27  *  In particular, GSS_Accept_sec_context is handled by an upcall
  28  * Data exchange is handled entirely within the kernel
  29  *  In particular, GSS_GetMIC, GSS_VerifyMIC, GSS_Seal, GSS_Unseal are in-kernel.
  30  * Context destruction is handled in-kernel
  31  *  GSS_Delete_sec_context is in-kernel
  32  *
  33  * Context creation is initiated by a RPCSEC_GSS_INIT request arriving.
  34  * The context handle and gss_token are used as a key into the rpcsec_init cache.
  35  * The content of this cache includes some of the outputs of GSS_Accept_sec_context,
  36  * being major_status, minor_status, context_handle, reply_token.
  37  * These are sent back to the client.
  38  * Sequence window management is handled by the kernel.  The window size if currently
  39  * a compile time constant.
  40  *
  41  * When user-space is happy that a context is established, it places an entry
  42  * in the rpcsec_context cache. The key for this cache is the context_handle.
  43  * The content includes:
  44  *   uid/gidlist - for determining access rights
  45  *   mechanism type
  46  *   mechanism specific information, such as a key
  47  *
  48  */
  49
  50 #define DEBUG_SUBSYSTEM S_SEC
  51 #include <linux/types.h>
  52 #include <linux/init.h>
  53 #include <linux/module.h>
  54 #include <linux/slab.h>
  55 #include <linux/hash.h>
  56 #include <linux/mutex.h>
  57 #include <linux/sunrpc/cache.h>
  58
  59 #include <obd.h>
  60 #include <obd_class.h>
  61 #include <obd_support.h>
  62 #include <lustre/lustre_idl.h>
  63 #include <lustre_net.h>
  64 #include <lustre_import.h>
  65 #include <lustre_sec.h>
  66
  67 #include "gss_err.h"
  68 #include "gss_internal.h"
  69 #include "gss_api.h"
  70
  71 #define GSS_SVC_UPCALL_TIMEOUT  (20)
  72
  73 static spinlock_t __ctx_index_lock;
  74 static __u64 __ctx_index;
  75
  76 __u64 gss_get_next_ctx_index(void)
  77 {
  78         __u64 idx;
  79
  80         spin_lock(&__ctx_index_lock);
  81         idx = __ctx_index++;
  82         spin_unlock(&__ctx_index_lock);
  83
  84         return idx;
  85 }
  86
  87 static inline unsigned long hash_mem(char *buf, int length, int bits)
  88 {
  89         unsigned long hash = 0;
  90         unsigned long l = 0;
  91         int len = 0;
  92         unsigned char c;
  93
  94         do {
  95                 if (len == length) {
  96                         c = (char) len;
  97                         len = -1;
  98                 } else
  99                         c = *buf++;
 100
 101                 l = (l << 8) | c;
 102                 len++;
 103
 104                 if ((len & (BITS_PER_LONG/8-1)) == 0)
 105                         hash = cfs_hash_long(hash^l, BITS_PER_LONG);
 106         } while (len);
 107
 108         return hash >> (BITS_PER_LONG - bits);
 109 }
 110
 111 /****************************************
 112  * rsi cache                        *
 113  ****************************************/
 114
 115 #define RSI_HASHBITS    (6)
 116 #define RSI_HASHMAX     (1 << RSI_HASHBITS)
 117 #define RSI_HASHMASK    (RSI_HASHMAX - 1)
 118
 119 struct rsi {
 120         struct cache_head       h;
 121         __u32              lustre_svc;
 122         __u64              nid;
 123         wait_queue_head_t            waitq;
 124         rawobj_t                in_handle, in_token;
 125         rawobj_t                out_handle, out_token;
 126         int                  major_status, minor_status;
 127 };
 128
 129 static struct cache_head *rsi_table[RSI_HASHMAX];
 130 static struct cache_detail rsi_cache;
 131 static struct rsi *rsi_update(struct rsi *new, struct rsi *old);
 132 static struct rsi *rsi_lookup(struct rsi *item);
 133
 134 static inline int rsi_hash(struct rsi *item)
 135 {
 136         return hash_mem((char *)item->in_handle.data, item->in_handle.len,
 137                         RSI_HASHBITS) ^
 138                hash_mem((char *)item->in_token.data, item->in_token.len,
 139                         RSI_HASHBITS);
 140 }
 141
 142 static inline int __rsi_match(struct rsi *item, struct rsi *tmp)
 143 {
 144         return (rawobj_equal(&item->in_handle, &tmp->in_handle) &&
 145                 rawobj_equal(&item->in_token, &tmp->in_token));
 146 }
 147
 148 static void rsi_free(struct rsi *rsi)
 149 {
 150         rawobj_free(&rsi->in_handle);
 151         rawobj_free(&rsi->in_token);
 152         rawobj_free(&rsi->out_handle);
 153         rawobj_free(&rsi->out_token);
 154 }
 155
 156 static void rsi_request(struct cache_detail *cd,
 157                         struct cache_head *h,
 158                         char **bpp, int *blen)
 159 {
 160         struct rsi *rsi = container_of(h, struct rsi, h);
 161         __u64 index = 0;
 162
 163         /* if in_handle is null, provide kernel suggestion */
 164         if (rsi->in_handle.len == 0)
 165                 index = gss_get_next_ctx_index();
 166
 167         qword_addhex(bpp, blen, (char *) &rsi->lustre_svc,
 168                      sizeof(rsi->lustre_svc));
 169         qword_addhex(bpp, blen, (char *) &rsi->nid, sizeof(rsi->nid));
 170         qword_addhex(bpp, blen, (char *) &index, sizeof(index));
 171         qword_addhex(bpp, blen, rsi->in_handle.data, rsi->in_handle.len);
 172         qword_addhex(bpp, blen, rsi->in_token.data, rsi->in_token.len);
 173         (*bpp)[-1] = '\n';
 174 }
 175
 176 static int rsi_upcall(struct cache_detail *cd, struct cache_head *h)
 177 {
 178         return sunrpc_cache_pipe_upcall(cd, h, rsi_request);
 179 }
 180
 181 static inline void __rsi_init(struct rsi *new, struct rsi *item)
 182 {
 183         new->out_handle = RAWOBJ_EMPTY;
 184         new->out_token = RAWOBJ_EMPTY;
 185
 186         new->in_handle = item->in_handle;
 187         item->in_handle = RAWOBJ_EMPTY;
 188         new->in_token = item->in_token;
 189         item->in_token = RAWOBJ_EMPTY;
 190
 191         new->lustre_svc = item->lustre_svc;
 192         new->nid = item->nid;
 193         init_waitqueue_head(&new->waitq);
 194 }
 195
 196 static inline void __rsi_update(struct rsi *new, struct rsi *item)
 197 {
 198         LASSERT(new->out_handle.len == 0);
 199         LASSERT(new->out_token.len == 0);
 200
 201         new->out_handle = item->out_handle;
 202         item->out_handle = RAWOBJ_EMPTY;
 203         new->out_token = item->out_token;
 204         item->out_token = RAWOBJ_EMPTY;
 205
 206         new->major_status = item->major_status;
 207         new->minor_status = item->minor_status;
 208 }
 209
 210 static void rsi_put(struct kref *ref)
 211 {
 212         struct rsi *rsi = container_of(ref, struct rsi, h.ref);
 213
 214         LASSERT(rsi->h.next == NULL);
 215         rsi_free(rsi);
 216         OBD_FREE_PTR(rsi);
 217 }
 218
 219 static int rsi_match(struct cache_head *a, struct cache_head *b)
 220 {
 221         struct rsi *item = container_of(a, struct rsi, h);
 222         struct rsi *tmp = container_of(b, struct rsi, h);
 223
 224         return __rsi_match(item, tmp);
 225 }
 226
 227 static void rsi_init(struct cache_head *cnew, struct cache_head *citem)
 228 {
 229         struct rsi *new = container_of(cnew, struct rsi, h);
 230         struct rsi *item = container_of(citem, struct rsi, h);
 231
 232         __rsi_init(new, item);
 233 }
 234
 235 static void update_rsi(struct cache_head *cnew, struct cache_head *citem)
 236 {
 237         struct rsi *new = container_of(cnew, struct rsi, h);
 238         struct rsi *item = container_of(citem, struct rsi, h);
 239
 240         __rsi_update(new, item);
 241 }
 242
 243 static struct cache_head *rsi_alloc(void)
 244 {
 245         struct rsi *rsi;
 246
 247         OBD_ALLOC_PTR(rsi);
 248         if (rsi)
 249                 return &rsi->h;
 250         else
 251                 return NULL;
 252 }
 253
 254 static int rsi_parse(struct cache_detail *cd, char *mesg, int mlen)
 255 {
 256         char       *buf = mesg;
 257         char       *ep;
 258         int          len;
 259         struct rsi      rsii, *rsip = NULL;
 260         time_t    expiry;
 261         int          status = -EINVAL;
 262
 263         memset(&rsii, 0, sizeof(rsii));
 264
 265         /* handle */
 266         len = qword_get(&mesg, buf, mlen);
 267         if (len < 0)
 268                 goto out;
 269         if (rawobj_alloc(&rsii.in_handle, buf, len)) {
 270                 status = -ENOMEM;
 271                 goto out;
 272         }
 273
 274         /* token */
 275         len = qword_get(&mesg, buf, mlen);
 276         if (len < 0)
 277                 goto out;
 278         if (rawobj_alloc(&rsii.in_token, buf, len)) {
 279                 status = -ENOMEM;
 280                 goto out;
 281         }
 282
 283         rsip = rsi_lookup(&rsii);
 284         if (!rsip)
 285                 goto out;
 286
 287         rsii.h.flags = 0;
 288         /* expiry */
 289         expiry = get_expiry(&mesg);
 290         if (expiry == 0)
 291                 goto out;
 292
 293         len = qword_get(&mesg, buf, mlen);
 294         if (len <= 0)
 295                 goto out;
 296
 297         /* major */
 298         rsii.major_status = simple_strtol(buf, &ep, 10);
 299         if (*ep)
 300                 goto out;
 301
 302         /* minor */
 303         len = qword_get(&mesg, buf, mlen);
 304         if (len <= 0)
 305                 goto out;
 306         rsii.minor_status = simple_strtol(buf, &ep, 10);
 307         if (*ep)
 308                 goto out;
 309
 310         /* out_handle */
 311         len = qword_get(&mesg, buf, mlen);
 312         if (len < 0)
 313                 goto out;
 314         if (rawobj_alloc(&rsii.out_handle, buf, len)) {
 315                 status = -ENOMEM;
 316                 goto out;
 317         }
 318
 319         /* out_token */
 320         len = qword_get(&mesg, buf, mlen);
 321         if (len < 0)
 322                 goto out;
 323         if (rawobj_alloc(&rsii.out_token, buf, len)) {
 324                 status = -ENOMEM;
 325                 goto out;
 326         }
 327
 328         rsii.h.expiry_time = expiry;
 329         rsip = rsi_update(&rsii, rsip);
 330         status = 0;
 331 out:
 332         rsi_free(&rsii);
 333         if (rsip) {
 334                 wake_up_all(&rsip->waitq);
 335                 cache_put(&rsip->h, &rsi_cache);
 336         } else {
 337                 status = -ENOMEM;
 338         }
 339
 340         if (status)
 341                 CERROR("rsi parse error %d\n", status);
 342         return status;
 343 }
 344
 345 static struct cache_detail rsi_cache = {
 346         .hash_size      = RSI_HASHMAX,
 347         .hash_table     = rsi_table,
 348         .name      = "auth.sptlrpc.init",
 349         .cache_put      = rsi_put,
 350         .cache_upcall   = rsi_upcall,
 351         .cache_parse    = rsi_parse,
 352         .match    = rsi_match,
 353         .init      = rsi_init,
 354         .update  = update_rsi,
 355         .alloc    = rsi_alloc,
 356 };
 357
 358 static struct rsi *rsi_lookup(struct rsi *item)
 359 {
 360         struct cache_head *ch;
 361         int hash = rsi_hash(item);
 362
 363         ch = sunrpc_cache_lookup(&rsi_cache, &item->h, hash);
 364         if (ch)
 365                 return container_of(ch, struct rsi, h);
 366         else
 367                 return NULL;
 368 }
 369
 370 static struct rsi *rsi_update(struct rsi *new, struct rsi *old)
 371 {
 372         struct cache_head *ch;
 373         int hash = rsi_hash(new);
 374
 375         ch = sunrpc_cache_update(&rsi_cache, &new->h, &old->h, hash);
 376         if (ch)
 377                 return container_of(ch, struct rsi, h);
 378         else
 379                 return NULL;
 380 }
 381
 382 /****************************************
 383  * rsc cache                        *
 384  ****************************************/
 385
 386 #define RSC_HASHBITS    (10)
 387 #define RSC_HASHMAX     (1 << RSC_HASHBITS)
 388 #define RSC_HASHMASK    (RSC_HASHMAX - 1)
 389
 390 struct rsc {
 391         struct cache_head       h;
 392         struct obd_device      *target;
 393         rawobj_t                handle;
 394         struct gss_svc_ctx      ctx;
 395 };
 396
 397 static struct cache_head *rsc_table[RSC_HASHMAX];
 398 static struct cache_detail rsc_cache;
 399 static struct rsc *rsc_update(struct rsc *new, struct rsc *old);
 400 static struct rsc *rsc_lookup(struct rsc *item);
 401
 402 static void rsc_free(struct rsc *rsci)
 403 {
 404         rawobj_free(&rsci->handle);
 405         rawobj_free(&rsci->ctx.gsc_rvs_hdl);
 406         lgss_delete_sec_context(&rsci->ctx.gsc_mechctx);
 407 }
 408
 409 static inline int rsc_hash(struct rsc *rsci)
 410 {
 411         return hash_mem((char *)rsci->handle.data,
 412                         rsci->handle.len, RSC_HASHBITS);
 413 }
 414
 415 static inline int __rsc_match(struct rsc *new, struct rsc *tmp)
 416 {
 417         return rawobj_equal(&new->handle, &tmp->handle);
 418 }
 419
 420 static inline void __rsc_init(struct rsc *new, struct rsc *tmp)
 421 {
 422         new->handle = tmp->handle;
 423         tmp->handle = RAWOBJ_EMPTY;
 424
 425         new->target = NULL;
 426         memset(&new->ctx, 0, sizeof(new->ctx));
 427         new->ctx.gsc_rvs_hdl = RAWOBJ_EMPTY;
 428 }
 429
 430 static inline void __rsc_update(struct rsc *new, struct rsc *tmp)
 431 {
 432         new->ctx = tmp->ctx;
 433         tmp->ctx.gsc_rvs_hdl = RAWOBJ_EMPTY;
 434         tmp->ctx.gsc_mechctx = NULL;
 435
 436         memset(&new->ctx.gsc_seqdata, 0, sizeof(new->ctx.gsc_seqdata));
 437         spin_lock_init(&new->ctx.gsc_seqdata.ssd_lock);
 438 }
 439
 440 static void rsc_put(struct kref *ref)
 441 {
 442         struct rsc *rsci = container_of(ref, struct rsc, h.ref);
 443
 444         LASSERT(rsci->h.next == NULL);
 445         rsc_free(rsci);
 446         OBD_FREE_PTR(rsci);
 447 }
 448
 449 static int rsc_match(struct cache_head *a, struct cache_head *b)
 450 {
 451         struct rsc *new = container_of(a, struct rsc, h);
 452         struct rsc *tmp = container_of(b, struct rsc, h);
 453
 454         return __rsc_match(new, tmp);
 455 }
 456
 457 static void rsc_init(struct cache_head *cnew, struct cache_head *ctmp)
 458 {
 459         struct rsc *new = container_of(cnew, struct rsc, h);
 460         struct rsc *tmp = container_of(ctmp, struct rsc, h);
 461
 462         __rsc_init(new, tmp);
 463 }
 464
 465 static void update_rsc(struct cache_head *cnew, struct cache_head *ctmp)
 466 {
 467         struct rsc *new = container_of(cnew, struct rsc, h);
 468         struct rsc *tmp = container_of(ctmp, struct rsc, h);
 469
 470         __rsc_update(new, tmp);
 471 }
 472
 473 static struct cache_head * rsc_alloc(void)
 474 {
 475         struct rsc *rsc;
 476
 477         OBD_ALLOC_PTR(rsc);
 478         if (rsc)
 479                 return &rsc->h;
 480         else
 481                 return NULL;
 482 }
 483
 484 static int rsc_parse(struct cache_detail *cd, char *mesg, int mlen)
 485 {
 486         char            *buf = mesg;
 487         int               len, rv, tmp_int;
 488         struct rsc         rsci, *rscp = NULL;
 489         time_t         expiry;
 490         int               status = -EINVAL;
 491         struct gss_api_mech *gm = NULL;
 492
 493         memset(&rsci, 0, sizeof(rsci));
 494
 495         /* context handle */
 496         len = qword_get(&mesg, buf, mlen);
 497         if (len < 0) goto out;
 498         status = -ENOMEM;
 499         if (rawobj_alloc(&rsci.handle, buf, len))
 500                 goto out;
 501
 502         rsci.h.flags = 0;
 503         /* expiry */
 504         expiry = get_expiry(&mesg);
 505         status = -EINVAL;
 506         if (expiry == 0)
 507                 goto out;
 508
 509         /* remote flag */
 510         rv = get_int(&mesg, &tmp_int);
 511         if (rv) {
 512                 CERROR("fail to get remote flag\n");
 513                 goto out;
 514         }
 515         rsci.ctx.gsc_remote = (tmp_int != 0);
 516
 517         /* root user flag */
 518         rv = get_int(&mesg, &tmp_int);
 519         if (rv) {
 520                 CERROR("fail to get oss user flag\n");
 521                 goto out;
 522         }
 523         rsci.ctx.gsc_usr_root = (tmp_int != 0);
 524
 525         /* mds user flag */
 526         rv = get_int(&mesg, &tmp_int);
 527         if (rv) {
 528                 CERROR("fail to get mds user flag\n");
 529                 goto out;
 530         }
 531         rsci.ctx.gsc_usr_mds = (tmp_int != 0);
 532
 533         /* oss user flag */
 534         rv = get_int(&mesg, &tmp_int);
 535         if (rv) {
 536                 CERROR("fail to get oss user flag\n");
 537                 goto out;
 538         }
 539         rsci.ctx.gsc_usr_oss = (tmp_int != 0);
 540
 541         /* mapped uid */
 542         rv = get_int(&mesg, (int *) &rsci.ctx.gsc_mapped_uid);
 543         if (rv) {
 544                 CERROR("fail to get mapped uid\n");
 545                 goto out;
 546         }
 547
 548         rscp = rsc_lookup(&rsci);
 549         if (!rscp)
 550                 goto out;
 551
 552         /* uid, or NEGATIVE */
 553         rv = get_int(&mesg, (int *) &rsci.ctx.gsc_uid);
 554         if (rv == -EINVAL)
 555                 goto out;
 556         if (rv == -ENOENT) {
 557                 CERROR("NOENT? set rsc entry negative\n");
 558                 set_bit(CACHE_NEGATIVE, &rsci.h.flags);
 559         } else {
 560                 rawobj_t tmp_buf;
 561                 unsigned long ctx_expiry;
 562
 563                 /* gid */
 564                 if (get_int(&mesg, (int *) &rsci.ctx.gsc_gid))
 565                         goto out;
 566
 567                 /* mech name */
 568                 len = qword_get(&mesg, buf, mlen);
 569                 if (len < 0)
 570                         goto out;
 571                 gm = lgss_name_to_mech(buf);
 572                 status = -EOPNOTSUPP;
 573                 if (!gm)
 574                         goto out;
 575
 576                 status = -EINVAL;
 577                 /* mech-specific data: */
 578                 len = qword_get(&mesg, buf, mlen);
 579                 if (len < 0)
 580                         goto out;
 581
 582                 tmp_buf.len = len;
 583                 tmp_buf.data = (unsigned char *)buf;
 584                 if (lgss_import_sec_context(&tmp_buf, gm,
 585                                             &rsci.ctx.gsc_mechctx))
 586                         goto out;
 587
 588                 /* currently the expiry time passed down from user-space
 589                  * is invalid, here we retrive it from mech. */
 590                 if (lgss_inquire_context(rsci.ctx.gsc_mechctx, &ctx_expiry)) {
 591                         CERROR("unable to get expire time, drop it\n");
 592                         goto out;
 593                 }
 594                 expiry = (time_t) ctx_expiry;
 595         }
 596
 597         rsci.h.expiry_time = expiry;
 598         rscp = rsc_update(&rsci, rscp);
 599         status = 0;
 600 out:
 601         if (gm)
 602                 lgss_mech_put(gm);
 603         rsc_free(&rsci);
 604         if (rscp)
 605                 cache_put(&rscp->h, &rsc_cache);
 606         else
 607                 status = -ENOMEM;
 608
 609         if (status)
 610                 CERROR("parse rsc error %d\n", status);
 611         return status;
 612 }
 613
 614 static struct cache_detail rsc_cache = {
 615         .hash_size      = RSC_HASHMAX,
 616         .hash_table     = rsc_table,
 617         .name      = "auth.sptlrpc.context",
 618         .cache_put      = rsc_put,
 619         .cache_parse    = rsc_parse,
 620         .match    = rsc_match,
 621         .init      = rsc_init,
 622         .update  = update_rsc,
 623         .alloc    = rsc_alloc,
 624 };
 625
 626 static struct rsc *rsc_lookup(struct rsc *item)
 627 {
 628         struct cache_head *ch;
 629         int             hash = rsc_hash(item);
 630
 631         ch = sunrpc_cache_lookup(&rsc_cache, &item->h, hash);
 632         if (ch)
 633                 return container_of(ch, struct rsc, h);
 634         else
 635                 return NULL;
 636 }
 637
 638 static struct rsc *rsc_update(struct rsc *new, struct rsc *old)
 639 {
 640         struct cache_head *ch;
 641         int             hash = rsc_hash(new);
 642
 643         ch = sunrpc_cache_update(&rsc_cache, &new->h, &old->h, hash);
 644         if (ch)
 645                 return container_of(ch, struct rsc, h);
 646         else
 647                 return NULL;
 648 }
 649
 650 #define COMPAT_RSC_PUT(item, cd)        cache_put((item), (cd))
 651
 652 /****************************************
 653  * rsc cache flush                    *
 654  ****************************************/
 655
 656 typedef int rsc_entry_match(struct rsc *rscp, long data);
 657
 658 static void rsc_flush(rsc_entry_match *match, long data)
 659 {
 660         struct cache_head **ch;
 661         struct rsc *rscp;
 662         int n;
 663
 664         write_lock(&rsc_cache.hash_lock);
 665         for (n = 0; n < RSC_HASHMAX; n++) {
 666                 for (ch = &rsc_cache.hash_table[n]; *ch;) {
 667                         rscp = container_of(*ch, struct rsc, h);
 668
 669                         if (!match(rscp, data)) {
 670                                 ch = &((*ch)->next);
 671                                 continue;
 672                         }
 673
 674                         /* it seems simply set NEGATIVE doesn't work */
 675                         *ch = (*ch)->next;
 676                         rscp->h.next = NULL;
 677                         cache_get(&rscp->h);
 678                         set_bit(CACHE_NEGATIVE, &rscp->h.flags);
 679                         COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
 680                         rsc_cache.entries--;
 681                 }
 682         }
 683         write_unlock(&rsc_cache.hash_lock);
 684 }
 685
 686 static int match_uid(struct rsc *rscp, long uid)
 687 {
 688         if ((int) uid == -1)
 689                 return 1;
 690         return ((int) rscp->ctx.gsc_uid == (int) uid);
 691 }
 692
 693 static int match_target(struct rsc *rscp, long target)
 694 {
 695         return (rscp->target == (struct obd_device *) target);
 696 }
 697
 698 static inline void rsc_flush_uid(int uid)
 699 {
 700         if (uid == -1)
 701                 CWARN("flush all gss contexts...\n");
 702
 703         rsc_flush(match_uid, (long) uid);
 704 }
 705
 706 static inline void rsc_flush_target(struct obd_device *target)
 707 {
 708         rsc_flush(match_target, (long) target);
 709 }
 710
 711 void gss_secsvc_flush(struct obd_device *target)
 712 {
 713         rsc_flush_target(target);
 714 }
 715 EXPORT_SYMBOL(gss_secsvc_flush);
 716
 717 static struct rsc *gss_svc_searchbyctx(rawobj_t *handle)
 718 {
 719         struct rsc  rsci;
 720         struct rsc *found;
 721
 722         memset(&rsci, 0, sizeof(rsci));
 723         if (rawobj_dup(&rsci.handle, handle))
 724                 return NULL;
 725
 726         found = rsc_lookup(&rsci);
 727         rsc_free(&rsci);
 728         if (!found)
 729                 return NULL;
 730         if (cache_check(&rsc_cache, &found->h, NULL))
 731                 return NULL;
 732         return found;
 733 }
 734
 735 int gss_svc_upcall_install_rvs_ctx(struct obd_import *imp,
 736                                    struct gss_sec *gsec,
 737                                    struct gss_cli_ctx *gctx)
 738 {
 739         struct rsc      rsci, *rscp = NULL;
 740         unsigned long   ctx_expiry;
 741         __u32      major;
 742         int          rc;
 743
 744         memset(&rsci, 0, sizeof(rsci));
 745
 746         if (rawobj_alloc(&rsci.handle, (char *) &gsec->gs_rvs_hdl,
 747                          sizeof(gsec->gs_rvs_hdl)))
 748                 GOTO(out, rc = -ENOMEM);
 749
 750         rscp = rsc_lookup(&rsci);
 751         if (rscp == NULL)
 752                 GOTO(out, rc = -ENOMEM);
 753
 754         major = lgss_copy_reverse_context(gctx->gc_mechctx,
 755                                           &rsci.ctx.gsc_mechctx);
 756         if (major != GSS_S_COMPLETE)
 757                 GOTO(out, rc = -ENOMEM);
 758
 759         if (lgss_inquire_context(rsci.ctx.gsc_mechctx, &ctx_expiry)) {
 760                 CERROR("unable to get expire time, drop it\n");
 761                 GOTO(out, rc = -EINVAL);
 762         }
 763         rsci.h.expiry_time = (time_t) ctx_expiry;
 764
 765         if (strcmp(imp->imp_obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0)
 766                 rsci.ctx.gsc_usr_mds = 1;
 767         else if (strcmp(imp->imp_obd->obd_type->typ_name, LUSTRE_OSC_NAME) == 0)
 768                 rsci.ctx.gsc_usr_oss = 1;
 769         else
 770                 rsci.ctx.gsc_usr_root = 1;
 771
 772         rscp = rsc_update(&rsci, rscp);
 773         if (rscp == NULL)
 774                 GOTO(out, rc = -ENOMEM);
 775
 776         rscp->target = imp->imp_obd;
 777         rawobj_dup(&gctx->gc_svc_handle, &rscp->handle);
 778
 779         CWARN("create reverse svc ctx %p to %s: idx "LPX64"\n",
 780               &rscp->ctx, obd2cli_tgt(imp->imp_obd), gsec->gs_rvs_hdl);
 781         rc = 0;
 782 out:
 783         if (rscp)
 784                 cache_put(&rscp->h, &rsc_cache);
 785         rsc_free(&rsci);
 786
 787         if (rc)
 788                 CERROR("create reverse svc ctx: idx "LPX64", rc %d\n",
 789                        gsec->gs_rvs_hdl, rc);
 790         return rc;
 791 }
 792
 793 int gss_svc_upcall_expire_rvs_ctx(rawobj_t *handle)
 794 {
 795         const cfs_time_t        expire = 20;
 796         struct rsc           *rscp;
 797
 798         rscp = gss_svc_searchbyctx(handle);
 799         if (rscp) {
 800                 CDEBUG(D_SEC, "reverse svcctx %p (rsc %p) expire soon\n",
 801                        &rscp->ctx, rscp);
 802
 803                 rscp->h.expiry_time = cfs_time_current_sec() + expire;
 804                 COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
 805         }
 806         return 0;
 807 }
 808
 809 int gss_svc_upcall_dup_handle(rawobj_t *handle, struct gss_svc_ctx *ctx)
 810 {
 811         struct rsc *rscp = container_of(ctx, struct rsc, ctx);
 812
 813         return rawobj_dup(handle, &rscp->handle);
 814 }
 815
 816 int gss_svc_upcall_update_sequence(rawobj_t *handle, __u32 seq)
 817 {
 818         struct rsc           *rscp;
 819
 820         rscp = gss_svc_searchbyctx(handle);
 821         if (rscp) {
 822                 CDEBUG(D_SEC, "reverse svcctx %p (rsc %p) update seq to %u\n",
 823                        &rscp->ctx, rscp, seq + 1);
 824
 825                 rscp->ctx.gsc_rvs_seq = seq + 1;
 826                 COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
 827         }
 828         return 0;
 829 }
 830
 831 static struct cache_deferred_req* cache_upcall_defer(struct cache_req *req)
 832 {
 833         return NULL;
 834 }
 835 static struct cache_req cache_upcall_chandle = { cache_upcall_defer };
 836
 837 int gss_svc_upcall_handle_init(struct ptlrpc_request *req,
 838                                struct gss_svc_reqctx *grctx,
 839                                struct gss_wire_ctx *gw,
 840                                struct obd_device *target,
 841                                __u32 lustre_svc,
 842                                rawobj_t *rvs_hdl,
 843                                rawobj_t *in_token)
 844 {
 845         struct ptlrpc_reply_state *rs;
 846         struct rsc              *rsci = NULL;
 847         struct rsi              *rsip = NULL, rsikey;
 848         wait_queue_t         wait;
 849         int                     replen = sizeof(struct ptlrpc_body);
 850         struct gss_rep_header     *rephdr;
 851         int                     first_check = 1;
 852         int                     rc = SECSVC_DROP;
 853
 854         memset(&rsikey, 0, sizeof(rsikey));
 855         rsikey.lustre_svc = lustre_svc;
 856         rsikey.nid = (__u64) req->rq_peer.nid;
 857
 858         /* duplicate context handle. for INIT it always 0 */
 859         if (rawobj_dup(&rsikey.in_handle, &gw->gw_handle)) {
 860                 CERROR("fail to dup context handle\n");
 861                 GOTO(out, rc);
 862         }
 863
 864         if (rawobj_dup(&rsikey.in_token, in_token)) {
 865                 CERROR("can't duplicate token\n");
 866                 rawobj_free(&rsikey.in_handle);
 867                 GOTO(out, rc);
 868         }
 869
 870         rsip = rsi_lookup(&rsikey);
 871         rsi_free(&rsikey);
 872         if (!rsip) {
 873                 CERROR("error in rsi_lookup.\n");
 874
 875                 if (!gss_pack_err_notify(req, GSS_S_FAILURE, 0))
 876                         rc = SECSVC_COMPLETE;
 877
 878                 GOTO(out, rc);
 879         }
 880
 881         cache_get(&rsip->h); /* take an extra ref */
 882         init_waitqueue_head(&rsip->waitq);
 883         init_waitqueue_entry_current(&wait);
 884         add_wait_queue(&rsip->waitq, &wait);
 885
 886 cache_check:
 887         /* Note each time cache_check() will drop a reference if return
 888          * non-zero. We hold an extra reference on initial rsip, but must
 889          * take care of following calls. */
 890         rc = cache_check(&rsi_cache, &rsip->h, &cache_upcall_chandle);
 891         switch (rc) {
 892         case -EAGAIN: {
 893                 int valid;
 894
 895                 if (first_check) {
 896                         first_check = 0;
 897
 898                         read_lock(&rsi_cache.hash_lock);
 899                         valid = test_bit(CACHE_VALID, &rsip->h.flags);
 900                         if (valid == 0)
 901                                 set_current_state(TASK_INTERRUPTIBLE);
 902                         read_unlock(&rsi_cache.hash_lock);
 903
 904                         if (valid == 0)
 905                                 schedule_timeout(GSS_SVC_UPCALL_TIMEOUT *
 906                                                      HZ);
 907
 908                         cache_get(&rsip->h);
 909                         goto cache_check;
 910                 }
 911                 CWARN("waited %ds timeout, drop\n", GSS_SVC_UPCALL_TIMEOUT);
 912                 break;
 913         }
 914         case -ENOENT:
 915                 CWARN("cache_check return ENOENT, drop\n");
 916                 break;
 917         case 0:
 918                 /* if not the first check, we have to release the extra
 919                  * reference we just added on it. */
 920                 if (!first_check)
 921                         cache_put(&rsip->h, &rsi_cache);
 922                 CDEBUG(D_SEC, "cache_check is good\n");
 923                 break;
 924         }
 925
 926         remove_wait_queue(&rsip->waitq, &wait);
 927         cache_put(&rsip->h, &rsi_cache);
 928
 929         if (rc)
 930                 GOTO(out, rc = SECSVC_DROP);
 931
 932         rc = SECSVC_DROP;
 933         rsci = gss_svc_searchbyctx(&rsip->out_handle);
 934         if (!rsci) {
 935                 CERROR("authentication failed\n");
 936
 937                 if (!gss_pack_err_notify(req, GSS_S_FAILURE, 0))
 938                         rc = SECSVC_COMPLETE;
 939
 940                 GOTO(out, rc);
 941         } else {
 942                 cache_get(&rsci->h);
 943                 grctx->src_ctx = &rsci->ctx;
 944         }
 945
 946         if (rawobj_dup(&rsci->ctx.gsc_rvs_hdl, rvs_hdl)) {
 947                 CERROR("failed duplicate reverse handle\n");
 948                 GOTO(out, rc);
 949         }
 950
 951         rsci->target = target;
 952
 953         CDEBUG(D_SEC, "server create rsc %p(%u->%s)\n",
 954                rsci, rsci->ctx.gsc_uid, libcfs_nid2str(req->rq_peer.nid));
 955
 956         if (rsip->out_handle.len > PTLRPC_GSS_MAX_HANDLE_SIZE) {
 957                 CERROR("handle size %u too large\n", rsip->out_handle.len);
 958                 GOTO(out, rc = SECSVC_DROP);
 959         }
 960
 961         grctx->src_init = 1;
 962         grctx->src_reserve_len = cfs_size_round4(rsip->out_token.len);
 963
 964         rc = lustre_pack_reply_v2(req, 1, &replen, NULL, 0);
 965         if (rc) {
 966                 CERROR("failed to pack reply: %d\n", rc);
 967                 GOTO(out, rc = SECSVC_DROP);
 968         }
 969
 970         rs = req->rq_reply_state;
 971         LASSERT(rs->rs_repbuf->lm_bufcount == 3);
 972         LASSERT(rs->rs_repbuf->lm_buflens[0] >=
 973                 sizeof(*rephdr) + rsip->out_handle.len);
 974         LASSERT(rs->rs_repbuf->lm_buflens[2] >= rsip->out_token.len);
 975
 976         rephdr = lustre_msg_buf(rs->rs_repbuf, 0, 0);
 977         rephdr->gh_version = PTLRPC_GSS_VERSION;
 978         rephdr->gh_flags = 0;
 979         rephdr->gh_proc = PTLRPC_GSS_PROC_ERR;
 980         rephdr->gh_major = rsip->major_status;
 981         rephdr->gh_minor = rsip->minor_status;
 982         rephdr->gh_seqwin = GSS_SEQ_WIN;
 983         rephdr->gh_handle.len = rsip->out_handle.len;
 984         memcpy(rephdr->gh_handle.data, rsip->out_handle.data,
 985                rsip->out_handle.len);
 986
 987         memcpy(lustre_msg_buf(rs->rs_repbuf, 2, 0), rsip->out_token.data,
 988                rsip->out_token.len);
 989
 990         rs->rs_repdata_len = lustre_shrink_msg(rs->rs_repbuf, 2,
 991                                                rsip->out_token.len, 0);
 992
 993         rc = SECSVC_OK;
 994
 995 out:
 996         /* it looks like here we should put rsip also, but this mess up
 997          * with NFS cache mgmt code... FIXME */
 998 #if 0
 999         if (rsip)
1000                 rsi_put(&rsip->h, &rsi_cache);
1001 #endif
1002
1003         if (rsci) {
1004                 /* if anything went wrong, we don't keep the context too */
1005                 if (rc != SECSVC_OK)
1006                         set_bit(CACHE_NEGATIVE, &rsci->h.flags);
1007                 else
1008                         CDEBUG(D_SEC, "create rsc with idx "LPX64"\n",
1009                                gss_handle_to_u64(&rsci->handle));
1010
1011                 COMPAT_RSC_PUT(&rsci->h, &rsc_cache);
1012         }
1013         return rc;
1014 }
1015
1016 struct gss_svc_ctx *gss_svc_upcall_get_ctx(struct ptlrpc_request *req,
1017                                            struct gss_wire_ctx *gw)
1018 {
1019         struct rsc *rsc;
1020
1021         rsc = gss_svc_searchbyctx(&gw->gw_handle);
1022         if (!rsc) {
1023                 CWARN("Invalid gss ctx idx "LPX64" from %s\n",
1024                       gss_handle_to_u64(&gw->gw_handle),
1025                       libcfs_nid2str(req->rq_peer.nid));
1026                 return NULL;
1027         }
1028
1029         return &rsc->ctx;
1030 }
1031
1032 void gss_svc_upcall_put_ctx(struct gss_svc_ctx *ctx)
1033 {
1034         struct rsc *rsc = container_of(ctx, struct rsc, ctx);
1035
1036         COMPAT_RSC_PUT(&rsc->h, &rsc_cache);
1037 }
1038
1039 void gss_svc_upcall_destroy_ctx(struct gss_svc_ctx *ctx)
1040 {
1041         struct rsc *rsc = container_of(ctx, struct rsc, ctx);
1042
1043         /* can't be found */
1044         set_bit(CACHE_NEGATIVE, &rsc->h.flags);
1045         /* to be removed at next scan */
1046         rsc->h.expiry_time = 1;
1047 }
1048
1049 int __init gss_init_svc_upcall(void)
1050 {
1051         int     i;
1052
1053         spin_lock_init(&__ctx_index_lock);
1054         /*
1055          * this helps reducing context index confliction. after server reboot,
1056          * conflicting request from clients might be filtered out by initial
1057          * sequence number checking, thus no chance to sent error notification
1058          * back to clients.
1059          */
1060         cfs_get_random_bytes(&__ctx_index, sizeof(__ctx_index));
1061
1062
1063         cache_register(&rsi_cache);
1064         cache_register(&rsc_cache);
1065
1066         /* FIXME this looks stupid. we intend to give lsvcgssd a chance to open
1067          * the init upcall channel, otherwise there's big chance that the first
1068          * upcall issued before the channel be opened thus nfsv4 cache code will
1069          * drop the request direclty, thus lead to unnecessary recovery time.
1070          * here we wait at miximum 1.5 seconds. */
1071         for (i = 0; i < 6; i++) {
1072                 if (atomic_read(&rsi_cache.readers) > 0)
1073                         break;
1074                 set_current_state(TASK_UNINTERRUPTIBLE);
1075                 LASSERT(HZ >= 4);
1076                 schedule_timeout(HZ / 4);
1077         }
1078
1079         if (atomic_read(&rsi_cache.readers) == 0)
1080                 CWARN("Init channel is not opened by lsvcgssd, following "
1081                       "request might be dropped until lsvcgssd is active\n");
1082
1083         return 0;
1084 }
1085
1086 void __exit gss_exit_svc_upcall(void)
1087 {
1088         cache_purge(&rsi_cache);
1089         cache_unregister(&rsi_cache);
1090
1091         cache_purge(&rsc_cache);
1092         cache_unregister(&rsc_cache);
1093 }