fs/nfsd/nfs4callback.c

   1 /*
   2  *  Copyright (c) 2001 The Regents of the University of Michigan.
   3  *  All rights reserved.
   4  *
   5  *  Kendrick Smith <kmsmith@umich.edu>
   6  *  Andy Adamson <andros@umich.edu>
   7  *
   8  *  Redistribution and use in source and binary forms, with or without
   9  *  modification, are permitted provided that the following conditions
  10  *  are met:
  11  *
  12  *  1. Redistributions of source code must retain the above copyright
  13  *     notice, this list of conditions and the following disclaimer.
  14  *  2. Redistributions in binary form must reproduce the above copyright
  15  *     notice, this list of conditions and the following disclaimer in the
  16  *     documentation and/or other materials provided with the distribution.
  17  *  3. Neither the name of the University nor the names of its
  18  *     contributors may be used to endorse or promote products derived
  19  *     from this software without specific prior written permission.
  20  *
  21  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <linux/sunrpc/clnt.h>
  35 #include <linux/sunrpc/xprt.h>
  36 #include <linux/sunrpc/svc_xprt.h>
  37 #include <linux/slab.h>
  38 #include "nfsd.h"
  39 #include "state.h"
  40 #include "netns.h"
  41 #include "xdr4cb.h"
  42
  43 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  44
  45 static void nfsd4_mark_cb_fault(struct nfs4_client *, int reason);
  46
  47 #define NFSPROC4_CB_NULL 0
  48 #define NFSPROC4_CB_COMPOUND 1
  49
  50 /* Index of predefined Linux callback client operations */
  51
  52 struct nfs4_cb_compound_hdr {
  53         /* args */
  54         u32             ident;  /* minorversion 0 only */
  55         u32             nops;
  56         __be32          *nops_p;
  57         u32             minorversion;
  58         /* res */
  59         int             status;
  60 };
  61
  62 /*
  63  * Handle decode buffer overflows out-of-line.
  64  */
  65 static void print_overflow_msg(const char *func, const struct xdr_stream *xdr)
  66 {
  67         dprintk("NFS: %s prematurely hit the end of our receive buffer. "
  68                 "Remaining buffer length is %tu words.\n",
  69                 func, xdr->end - xdr->p);
  70 }
  71
  72 static __be32 *xdr_encode_empty_array(__be32 *p)
  73 {
  74         *p++ = xdr_zero;
  75         return p;
  76 }
  77
  78 /*
  79  * Encode/decode NFSv4 CB basic data types
  80  *
  81  * Basic NFSv4 callback data types are defined in section 15 of RFC
  82  * 3530: "Network File System (NFS) version 4 Protocol" and section
  83  * 20 of RFC 5661: "Network File System (NFS) Version 4 Minor Version
  84  * 1 Protocol"
  85  */
  86
  87 /*
  88  *      nfs_cb_opnum4
  89  *
  90  *      enum nfs_cb_opnum4 {
  91  *              OP_CB_GETATTR           = 3,
  92  *                ...
  93  *      };
  94  */
  95 enum nfs_cb_opnum4 {
  96         OP_CB_GETATTR                   = 3,
  97         OP_CB_RECALL                    = 4,
  98         OP_CB_LAYOUTRECALL              = 5,
  99         OP_CB_NOTIFY                    = 6,
 100         OP_CB_PUSH_DELEG                = 7,
 101         OP_CB_RECALL_ANY                = 8,
 102         OP_CB_RECALLABLE_OBJ_AVAIL      = 9,
 103         OP_CB_RECALL_SLOT               = 10,
 104         OP_CB_SEQUENCE                  = 11,
 105         OP_CB_WANTS_CANCELLED           = 12,
 106         OP_CB_NOTIFY_LOCK               = 13,
 107         OP_CB_NOTIFY_DEVICEID           = 14,
 108         OP_CB_ILLEGAL                   = 10044
 109 };
 110
 111 static void encode_nfs_cb_opnum4(struct xdr_stream *xdr, enum nfs_cb_opnum4 op)
 112 {
 113         __be32 *p;
 114
 115         p = xdr_reserve_space(xdr, 4);
 116         *p = cpu_to_be32(op);
 117 }
 118
 119 /*
 120  * nfs_fh4
 121  *
 122  *      typedef opaque nfs_fh4<NFS4_FHSIZE>;
 123  */
 124 static void encode_nfs_fh4(struct xdr_stream *xdr, const struct knfsd_fh *fh)
 125 {
 126         u32 length = fh->fh_size;
 127         __be32 *p;
 128
 129         BUG_ON(length > NFS4_FHSIZE);
 130         p = xdr_reserve_space(xdr, 4 + length);
 131         xdr_encode_opaque(p, &fh->fh_base, length);
 132 }
 133
 134 /*
 135  * stateid4
 136  *
 137  *      struct stateid4 {
 138  *              uint32_t        seqid;
 139  *              opaque          other[12];
 140  *      };
 141  */
 142 static void encode_stateid4(struct xdr_stream *xdr, const stateid_t *sid)
 143 {
 144         __be32 *p;
 145
 146         p = xdr_reserve_space(xdr, NFS4_STATEID_SIZE);
 147         *p++ = cpu_to_be32(sid->si_generation);
 148         xdr_encode_opaque_fixed(p, &sid->si_opaque, NFS4_STATEID_OTHER_SIZE);
 149 }
 150
 151 /*
 152  * sessionid4
 153  *
 154  *      typedef opaque sessionid4[NFS4_SESSIONID_SIZE];
 155  */
 156 static void encode_sessionid4(struct xdr_stream *xdr,
 157                               const struct nfsd4_session *session)
 158 {
 159         __be32 *p;
 160
 161         p = xdr_reserve_space(xdr, NFS4_MAX_SESSIONID_LEN);
 162         xdr_encode_opaque_fixed(p, session->se_sessionid.data,
 163                                         NFS4_MAX_SESSIONID_LEN);
 164 }
 165
 166 /*
 167  * nfsstat4
 168  */
 169 static const struct {
 170         int stat;
 171         int errno;
 172 } nfs_cb_errtbl[] = {
 173         { NFS4_OK,              0               },
 174         { NFS4ERR_PERM,         -EPERM          },
 175         { NFS4ERR_NOENT,        -ENOENT         },
 176         { NFS4ERR_IO,           -EIO            },
 177         { NFS4ERR_NXIO,         -ENXIO          },
 178         { NFS4ERR_ACCESS,       -EACCES         },
 179         { NFS4ERR_EXIST,        -EEXIST         },
 180         { NFS4ERR_XDEV,         -EXDEV          },
 181         { NFS4ERR_NOTDIR,       -ENOTDIR        },
 182         { NFS4ERR_ISDIR,        -EISDIR         },
 183         { NFS4ERR_INVAL,        -EINVAL         },
 184         { NFS4ERR_FBIG,         -EFBIG          },
 185         { NFS4ERR_NOSPC,        -ENOSPC         },
 186         { NFS4ERR_ROFS,         -EROFS          },
 187         { NFS4ERR_MLINK,        -EMLINK         },
 188         { NFS4ERR_NAMETOOLONG,  -ENAMETOOLONG   },
 189         { NFS4ERR_NOTEMPTY,     -ENOTEMPTY      },
 190         { NFS4ERR_DQUOT,        -EDQUOT         },
 191         { NFS4ERR_STALE,        -ESTALE         },
 192         { NFS4ERR_BADHANDLE,    -EBADHANDLE     },
 193         { NFS4ERR_BAD_COOKIE,   -EBADCOOKIE     },
 194         { NFS4ERR_NOTSUPP,      -ENOTSUPP       },
 195         { NFS4ERR_TOOSMALL,     -ETOOSMALL      },
 196         { NFS4ERR_SERVERFAULT,  -ESERVERFAULT   },
 197         { NFS4ERR_BADTYPE,      -EBADTYPE       },
 198         { NFS4ERR_LOCKED,       -EAGAIN         },
 199         { NFS4ERR_RESOURCE,     -EREMOTEIO      },
 200         { NFS4ERR_SYMLINK,      -ELOOP          },
 201         { NFS4ERR_OP_ILLEGAL,   -EOPNOTSUPP     },
 202         { NFS4ERR_DEADLOCK,     -EDEADLK        },
 203         { -1,                   -EIO            }
 204 };
 205
 206 /*
 207  * If we cannot translate the error, the recovery routines should
 208  * handle it.
 209  *
 210  * Note: remaining NFSv4 error codes have values > 10000, so should
 211  * not conflict with native Linux error codes.
 212  */
 213 static int nfs_cb_stat_to_errno(int status)
 214 {
 215         int i;
 216
 217         for (i = 0; nfs_cb_errtbl[i].stat != -1; i++) {
 218                 if (nfs_cb_errtbl[i].stat == status)
 219                         return nfs_cb_errtbl[i].errno;
 220         }
 221
 222         dprintk("NFSD: Unrecognized NFS CB status value: %u\n", status);
 223         return -status;
 224 }
 225
 226 static int decode_cb_op_status(struct xdr_stream *xdr, enum nfs_opnum4 expected,
 227                                int *status)
 228 {
 229         __be32 *p;
 230         u32 op;
 231
 232         p = xdr_inline_decode(xdr, 4 + 4);
 233         if (unlikely(p == NULL))
 234                 goto out_overflow;
 235         op = be32_to_cpup(p++);
 236         if (unlikely(op != expected))
 237                 goto out_unexpected;
 238         *status = nfs_cb_stat_to_errno(be32_to_cpup(p));
 239         return 0;
 240 out_overflow:
 241         print_overflow_msg(__func__, xdr);
 242         return -EIO;
 243 out_unexpected:
 244         dprintk("NFSD: Callback server returned operation %d but "
 245                 "we issued a request for %d\n", op, expected);
 246         return -EIO;
 247 }
 248
 249 /*
 250  * CB_COMPOUND4args
 251  *
 252  *      struct CB_COMPOUND4args {
 253  *              utf8str_cs      tag;
 254  *              uint32_t        minorversion;
 255  *              uint32_t        callback_ident;
 256  *              nfs_cb_argop4   argarray<>;
 257  *      };
 258 */
 259 static void encode_cb_compound4args(struct xdr_stream *xdr,
 260                                     struct nfs4_cb_compound_hdr *hdr)
 261 {
 262         __be32 * p;
 263
 264         p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4);
 265         p = xdr_encode_empty_array(p);          /* empty tag */
 266         *p++ = cpu_to_be32(hdr->minorversion);
 267         *p++ = cpu_to_be32(hdr->ident);
 268
 269         hdr->nops_p = p;
 270         *p = cpu_to_be32(hdr->nops);            /* argarray element count */
 271 }
 272
 273 /*
 274  * Update argarray element count
 275  */
 276 static void encode_cb_nops(struct nfs4_cb_compound_hdr *hdr)
 277 {
 278         BUG_ON(hdr->nops > NFS4_MAX_BACK_CHANNEL_OPS);
 279         *hdr->nops_p = cpu_to_be32(hdr->nops);
 280 }
 281
 282 /*
 283  * CB_COMPOUND4res
 284  *
 285  *      struct CB_COMPOUND4res {
 286  *              nfsstat4        status;
 287  *              utf8str_cs      tag;
 288  *              nfs_cb_resop4   resarray<>;
 289  *      };
 290  */
 291 static int decode_cb_compound4res(struct xdr_stream *xdr,
 292                                   struct nfs4_cb_compound_hdr *hdr)
 293 {
 294         u32 length;
 295         __be32 *p;
 296
 297         p = xdr_inline_decode(xdr, 4 + 4);
 298         if (unlikely(p == NULL))
 299                 goto out_overflow;
 300         hdr->status = be32_to_cpup(p++);
 301         /* Ignore the tag */
 302         length = be32_to_cpup(p++);
 303         p = xdr_inline_decode(xdr, length + 4);
 304         if (unlikely(p == NULL))
 305                 goto out_overflow;
 306         hdr->nops = be32_to_cpup(p);
 307         return 0;
 308 out_overflow:
 309         print_overflow_msg(__func__, xdr);
 310         return -EIO;
 311 }
 312
 313 /*
 314  * CB_RECALL4args
 315  *
 316  *      struct CB_RECALL4args {
 317  *              stateid4        stateid;
 318  *              bool            truncate;
 319  *              nfs_fh4         fh;
 320  *      };
 321  */
 322 static void encode_cb_recall4args(struct xdr_stream *xdr,
 323                                   const struct nfs4_delegation *dp,
 324                                   struct nfs4_cb_compound_hdr *hdr)
 325 {
 326         __be32 *p;
 327
 328         encode_nfs_cb_opnum4(xdr, OP_CB_RECALL);
 329         encode_stateid4(xdr, &dp->dl_stid.sc_stateid);
 330
 331         p = xdr_reserve_space(xdr, 4);
 332         *p++ = xdr_zero;                        /* truncate */
 333
 334         encode_nfs_fh4(xdr, &dp->dl_stid.sc_file->fi_fhandle);
 335
 336         hdr->nops++;
 337 }
 338
 339 /*
 340  * CB_SEQUENCE4args
 341  *
 342  *      struct CB_SEQUENCE4args {
 343  *              sessionid4              csa_sessionid;
 344  *              sequenceid4             csa_sequenceid;
 345  *              slotid4                 csa_slotid;
 346  *              slotid4                 csa_highest_slotid;
 347  *              bool                    csa_cachethis;
 348  *              referring_call_list4    csa_referring_call_lists<>;
 349  *      };
 350  */
 351 static void encode_cb_sequence4args(struct xdr_stream *xdr,
 352                                     const struct nfsd4_callback *cb,
 353                                     struct nfs4_cb_compound_hdr *hdr)
 354 {
 355         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
 356         __be32 *p;
 357
 358         if (hdr->minorversion == 0)
 359                 return;
 360
 361         encode_nfs_cb_opnum4(xdr, OP_CB_SEQUENCE);
 362         encode_sessionid4(xdr, session);
 363
 364         p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4 + 4);
 365         *p++ = cpu_to_be32(session->se_cb_seq_nr);      /* csa_sequenceid */
 366         *p++ = xdr_zero;                        /* csa_slotid */
 367         *p++ = xdr_zero;                        /* csa_highest_slotid */
 368         *p++ = xdr_zero;                        /* csa_cachethis */
 369         xdr_encode_empty_array(p);              /* csa_referring_call_lists */
 370
 371         hdr->nops++;
 372 }
 373
 374 /*
 375  * CB_SEQUENCE4resok
 376  *
 377  *      struct CB_SEQUENCE4resok {
 378  *              sessionid4      csr_sessionid;
 379  *              sequenceid4     csr_sequenceid;
 380  *              slotid4         csr_slotid;
 381  *              slotid4         csr_highest_slotid;
 382  *              slotid4         csr_target_highest_slotid;
 383  *      };
 384  *
 385  *      union CB_SEQUENCE4res switch (nfsstat4 csr_status) {
 386  *      case NFS4_OK:
 387  *              CB_SEQUENCE4resok       csr_resok4;
 388  *      default:
 389  *              void;
 390  *      };
 391  *
 392  * Our current back channel implmentation supports a single backchannel
 393  * with a single slot.
 394  */
 395 static int decode_cb_sequence4resok(struct xdr_stream *xdr,
 396                                     struct nfsd4_callback *cb)
 397 {
 398         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
 399         struct nfs4_sessionid id;
 400         int status;
 401         __be32 *p;
 402         u32 dummy;
 403
 404         status = -ESERVERFAULT;
 405
 406         /*
 407          * If the server returns different values for sessionID, slotID or
 408          * sequence number, the server is looney tunes.
 409          */
 410         p = xdr_inline_decode(xdr, NFS4_MAX_SESSIONID_LEN + 4 + 4 + 4 + 4);
 411         if (unlikely(p == NULL))
 412                 goto out_overflow;
 413         memcpy(id.data, p, NFS4_MAX_SESSIONID_LEN);
 414         if (memcmp(id.data, session->se_sessionid.data,
 415                                         NFS4_MAX_SESSIONID_LEN) != 0) {
 416                 dprintk("NFS: %s Invalid session id\n", __func__);
 417                 goto out;
 418         }
 419         p += XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN);
 420
 421         dummy = be32_to_cpup(p++);
 422         if (dummy != session->se_cb_seq_nr) {
 423                 dprintk("NFS: %s Invalid sequence number\n", __func__);
 424                 goto out;
 425         }
 426
 427         dummy = be32_to_cpup(p++);
 428         if (dummy != 0) {
 429                 dprintk("NFS: %s Invalid slotid\n", __func__);
 430                 goto out;
 431         }
 432
 433         /*
 434          * FIXME: process highest slotid and target highest slotid
 435          */
 436         status = 0;
 437 out:
 438         cb->cb_seq_status = status;
 439         return status;
 440 out_overflow:
 441         print_overflow_msg(__func__, xdr);
 442         status = -EIO;
 443         goto out;
 444 }
 445
 446 static int decode_cb_sequence4res(struct xdr_stream *xdr,
 447                                   struct nfsd4_callback *cb)
 448 {
 449         int status;
 450
 451         if (cb->cb_clp->cl_minorversion == 0)
 452                 return 0;
 453
 454         status = decode_cb_op_status(xdr, OP_CB_SEQUENCE, &cb->cb_seq_status);
 455         if (unlikely(status || cb->cb_seq_status))
 456                 return status;
 457
 458         return decode_cb_sequence4resok(xdr, cb);
 459 }
 460
 461 /*
 462  * NFSv4.0 and NFSv4.1 XDR encode functions
 463  *
 464  * NFSv4.0 callback argument types are defined in section 15 of RFC
 465  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
 466  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
 467  * Protocol".
 468  */
 469
 470 /*
 471  * NB: Without this zero space reservation, callbacks over krb5p fail
 472  */
 473 static void nfs4_xdr_enc_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
 474                                  void *__unused)
 475 {
 476         xdr_reserve_space(xdr, 0);
 477 }
 478
 479 /*
 480  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
 481  */
 482 static void nfs4_xdr_enc_cb_recall(struct rpc_rqst *req, struct xdr_stream *xdr,
 483                                    const struct nfsd4_callback *cb)
 484 {
 485         const struct nfs4_delegation *dp = cb_to_delegation(cb);
 486         struct nfs4_cb_compound_hdr hdr = {
 487                 .ident = cb->cb_clp->cl_cb_ident,
 488                 .minorversion = cb->cb_clp->cl_minorversion,
 489         };
 490
 491         encode_cb_compound4args(xdr, &hdr);
 492         encode_cb_sequence4args(xdr, cb, &hdr);
 493         encode_cb_recall4args(xdr, dp, &hdr);
 494         encode_cb_nops(&hdr);
 495 }
 496
 497
 498 /*
 499  * NFSv4.0 and NFSv4.1 XDR decode functions
 500  *
 501  * NFSv4.0 callback result types are defined in section 15 of RFC
 502  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
 503  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
 504  * Protocol".
 505  */
 506
 507 static int nfs4_xdr_dec_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
 508                                 void *__unused)
 509 {
 510         return 0;
 511 }
 512
 513 /*
 514  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
 515  */
 516 static int nfs4_xdr_dec_cb_recall(struct rpc_rqst *rqstp,
 517                                   struct xdr_stream *xdr,
 518                                   struct nfsd4_callback *cb)
 519 {
 520         struct nfs4_cb_compound_hdr hdr;
 521         int status;
 522
 523         status = decode_cb_compound4res(xdr, &hdr);
 524         if (unlikely(status))
 525                 return status;
 526
 527         if (cb != NULL) {
 528                 status = decode_cb_sequence4res(xdr, cb);
 529                 if (unlikely(status || cb->cb_seq_status))
 530                         return status;
 531         }
 532
 533         return decode_cb_op_status(xdr, OP_CB_RECALL, &cb->cb_status);
 534 }
 535
 536 #ifdef CONFIG_NFSD_PNFS
 537 /*
 538  * CB_LAYOUTRECALL4args
 539  *
 540  *      struct layoutrecall_file4 {
 541  *              nfs_fh4         lor_fh;
 542  *              offset4         lor_offset;
 543  *              length4         lor_length;
 544  *              stateid4        lor_stateid;
 545  *      };
 546  *
 547  *      union layoutrecall4 switch(layoutrecall_type4 lor_recalltype) {
 548  *      case LAYOUTRECALL4_FILE:
 549  *              layoutrecall_file4 lor_layout;
 550  *      case LAYOUTRECALL4_FSID:
 551  *              fsid4              lor_fsid;
 552  *      case LAYOUTRECALL4_ALL:
 553  *              void;
 554  *      };
 555  *
 556  *      struct CB_LAYOUTRECALL4args {
 557  *              layouttype4             clora_type;
 558  *              layoutiomode4           clora_iomode;
 559  *              bool                    clora_changed;
 560  *              layoutrecall4           clora_recall;
 561  *      };
 562  */
 563 static void encode_cb_layout4args(struct xdr_stream *xdr,
 564                                   const struct nfs4_layout_stateid *ls,
 565                                   struct nfs4_cb_compound_hdr *hdr)
 566 {
 567         __be32 *p;
 568
 569         BUG_ON(hdr->minorversion == 0);
 570
 571         p = xdr_reserve_space(xdr, 5 * 4);
 572         *p++ = cpu_to_be32(OP_CB_LAYOUTRECALL);
 573         *p++ = cpu_to_be32(ls->ls_layout_type);
 574         *p++ = cpu_to_be32(IOMODE_ANY);
 575         *p++ = cpu_to_be32(1);
 576         *p = cpu_to_be32(RETURN_FILE);
 577
 578         encode_nfs_fh4(xdr, &ls->ls_stid.sc_file->fi_fhandle);
 579
 580         p = xdr_reserve_space(xdr, 2 * 8);
 581         p = xdr_encode_hyper(p, 0);
 582         xdr_encode_hyper(p, NFS4_MAX_UINT64);
 583
 584         encode_stateid4(xdr, &ls->ls_recall_sid);
 585
 586         hdr->nops++;
 587 }
 588
 589 static void nfs4_xdr_enc_cb_layout(struct rpc_rqst *req,
 590                                    struct xdr_stream *xdr,
 591                                    const struct nfsd4_callback *cb)
 592 {
 593         const struct nfs4_layout_stateid *ls =
 594                 container_of(cb, struct nfs4_layout_stateid, ls_recall);
 595         struct nfs4_cb_compound_hdr hdr = {
 596                 .ident = 0,
 597                 .minorversion = cb->cb_clp->cl_minorversion,
 598         };
 599
 600         encode_cb_compound4args(xdr, &hdr);
 601         encode_cb_sequence4args(xdr, cb, &hdr);
 602         encode_cb_layout4args(xdr, ls, &hdr);
 603         encode_cb_nops(&hdr);
 604 }
 605
 606 static int nfs4_xdr_dec_cb_layout(struct rpc_rqst *rqstp,
 607                                   struct xdr_stream *xdr,
 608                                   struct nfsd4_callback *cb)
 609 {
 610         struct nfs4_cb_compound_hdr hdr;
 611         int status;
 612
 613         status = decode_cb_compound4res(xdr, &hdr);
 614         if (unlikely(status))
 615                 return status;
 616
 617         if (cb) {
 618                 status = decode_cb_sequence4res(xdr, cb);
 619                 if (unlikely(status || cb->cb_seq_status))
 620                         return status;
 621         }
 622         return decode_cb_op_status(xdr, OP_CB_LAYOUTRECALL, &cb->cb_status);
 623 }
 624 #endif /* CONFIG_NFSD_PNFS */
 625
 626 static void encode_stateowner(struct xdr_stream *xdr, struct nfs4_stateowner *so)
 627 {
 628         __be32  *p;
 629
 630         p = xdr_reserve_space(xdr, 8 + 4 + so->so_owner.len);
 631         p = xdr_encode_opaque_fixed(p, &so->so_client->cl_clientid, 8);
 632         xdr_encode_opaque(p, so->so_owner.data, so->so_owner.len);
 633 }
 634
 635 static void nfs4_xdr_enc_cb_notify_lock(struct rpc_rqst *req,
 636                                         struct xdr_stream *xdr,
 637                                         const struct nfsd4_callback *cb)
 638 {
 639         const struct nfsd4_blocked_lock *nbl =
 640                 container_of(cb, struct nfsd4_blocked_lock, nbl_cb);
 641         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)nbl->nbl_lock.fl_owner;
 642         struct nfs4_cb_compound_hdr hdr = {
 643                 .ident = 0,
 644                 .minorversion = cb->cb_clp->cl_minorversion,
 645         };
 646
 647         __be32 *p;
 648
 649         BUG_ON(hdr.minorversion == 0);
 650
 651         encode_cb_compound4args(xdr, &hdr);
 652         encode_cb_sequence4args(xdr, cb, &hdr);
 653
 654         p = xdr_reserve_space(xdr, 4);
 655         *p = cpu_to_be32(OP_CB_NOTIFY_LOCK);
 656         encode_nfs_fh4(xdr, &nbl->nbl_fh);
 657         encode_stateowner(xdr, &lo->lo_owner);
 658         hdr.nops++;
 659
 660         encode_cb_nops(&hdr);
 661 }
 662
 663 static int nfs4_xdr_dec_cb_notify_lock(struct rpc_rqst *rqstp,
 664                                         struct xdr_stream *xdr,
 665                                         struct nfsd4_callback *cb)
 666 {
 667         struct nfs4_cb_compound_hdr hdr;
 668         int status;
 669
 670         status = decode_cb_compound4res(xdr, &hdr);
 671         if (unlikely(status))
 672                 return status;
 673
 674         if (cb) {
 675                 status = decode_cb_sequence4res(xdr, cb);
 676                 if (unlikely(status || cb->cb_seq_status))
 677                         return status;
 678         }
 679         return decode_cb_op_status(xdr, OP_CB_NOTIFY_LOCK, &cb->cb_status);
 680 }
 681
 682 /*
 683  * RPC procedure tables
 684  */
 685 #define PROC(proc, call, argtype, restype)                              \
 686 [NFSPROC4_CLNT_##proc] = {                                              \
 687         .p_proc    = NFSPROC4_CB_##call,                                \
 688         .p_encode  = (kxdreproc_t)nfs4_xdr_enc_##argtype,               \
 689         .p_decode  = (kxdrdproc_t)nfs4_xdr_dec_##restype,               \
 690         .p_arglen  = NFS4_enc_##argtype##_sz,                           \
 691         .p_replen  = NFS4_dec_##restype##_sz,                           \
 692         .p_statidx = NFSPROC4_CB_##call,                                \
 693         .p_name    = #proc,                                             \
 694 }
 695
 696 static struct rpc_procinfo nfs4_cb_procedures[] = {
 697         PROC(CB_NULL,   NULL,           cb_null,        cb_null),
 698         PROC(CB_RECALL, COMPOUND,       cb_recall,      cb_recall),
 699 #ifdef CONFIG_NFSD_PNFS
 700         PROC(CB_LAYOUT, COMPOUND,       cb_layout,      cb_layout),
 701 #endif
 702         PROC(CB_NOTIFY_LOCK,    COMPOUND,       cb_notify_lock, cb_notify_lock),
 703 };
 704
 705 static struct rpc_version nfs_cb_version4 = {
 706 /*
 707  * Note on the callback rpc program version number: despite language in rfc
 708  * 5661 section 18.36.3 requiring servers to use 4 in this field, the
 709  * official xdr descriptions for both 4.0 and 4.1 specify version 1, and
 710  * in practice that appears to be what implementations use.  The section
 711  * 18.36.3 language is expected to be fixed in an erratum.
 712  */
 713         .number                 = 1,
 714         .nrprocs                = ARRAY_SIZE(nfs4_cb_procedures),
 715         .procs                  = nfs4_cb_procedures
 716 };
 717
 718 static const struct rpc_version *nfs_cb_version[] = {
 719         &nfs_cb_version4,
 720 };
 721
 722 static const struct rpc_program cb_program;
 723
 724 static struct rpc_stat cb_stats = {
 725         .program                = &cb_program
 726 };
 727
 728 #define NFS4_CALLBACK 0x40000000
 729 static const struct rpc_program cb_program = {
 730         .name                   = "nfs4_cb",
 731         .number                 = NFS4_CALLBACK,
 732         .nrvers                 = ARRAY_SIZE(nfs_cb_version),
 733         .version                = nfs_cb_version,
 734         .stats                  = &cb_stats,
 735         .pipe_dir_name          = "nfsd4_cb",
 736 };
 737
 738 static int max_cb_time(struct net *net)
 739 {
 740         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 741         return max(nn->nfsd4_lease/10, (time_t)1) * HZ;
 742 }
 743
 744 static struct rpc_cred *callback_cred;
 745
 746 int set_callback_cred(void)
 747 {
 748         if (callback_cred)
 749                 return 0;
 750         callback_cred = rpc_lookup_machine_cred("nfs");
 751         if (!callback_cred)
 752                 return -ENOMEM;
 753         return 0;
 754 }
 755
 756 static struct rpc_cred *get_backchannel_cred(struct nfs4_client *clp, struct rpc_clnt *client, struct nfsd4_session *ses)
 757 {
 758         if (clp->cl_minorversion == 0) {
 759                 return get_rpccred(callback_cred);
 760         } else {
 761                 struct rpc_auth *auth = client->cl_auth;
 762                 struct auth_cred acred = {};
 763
 764                 acred.uid = ses->se_cb_sec.uid;
 765                 acred.gid = ses->se_cb_sec.gid;
 766                 return auth->au_ops->lookup_cred(client->cl_auth, &acred, 0);
 767         }
 768 }
 769
 770 static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn, struct nfsd4_session *ses)
 771 {
 772         int maxtime = max_cb_time(clp->net);
 773         struct rpc_timeout      timeparms = {
 774                 .to_initval     = maxtime,
 775                 .to_retries     = 0,
 776                 .to_maxval      = maxtime,
 777         };
 778         struct rpc_create_args args = {
 779                 .net            = clp->net,
 780                 .address        = (struct sockaddr *) &conn->cb_addr,
 781                 .addrsize       = conn->cb_addrlen,
 782                 .saddress       = (struct sockaddr *) &conn->cb_saddr,
 783                 .timeout        = &timeparms,
 784                 .program        = &cb_program,
 785                 .version        = 0,
 786                 .flags          = (RPC_CLNT_CREATE_NOPING | RPC_CLNT_CREATE_QUIET),
 787         };
 788         struct rpc_clnt *client;
 789         struct rpc_cred *cred;
 790
 791         if (clp->cl_minorversion == 0) {
 792                 if (!clp->cl_cred.cr_principal &&
 793                                 (clp->cl_cred.cr_flavor >= RPC_AUTH_GSS_KRB5))
 794                         return -EINVAL;
 795                 args.client_name = clp->cl_cred.cr_principal;
 796                 args.prognumber = conn->cb_prog;
 797                 args.protocol = XPRT_TRANSPORT_TCP;
 798                 args.authflavor = clp->cl_cred.cr_flavor;
 799                 clp->cl_cb_ident = conn->cb_ident;
 800         } else {
 801                 if (!conn->cb_xprt)
 802                         return -EINVAL;
 803                 clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
 804                 clp->cl_cb_session = ses;
 805                 args.bc_xprt = conn->cb_xprt;
 806                 args.prognumber = clp->cl_cb_session->se_cb_prog;
 807                 args.protocol = conn->cb_xprt->xpt_class->xcl_ident |
 808                                 XPRT_TRANSPORT_BC;
 809                 args.authflavor = ses->se_cb_sec.flavor;
 810         }
 811         /* Create RPC client */
 812         client = rpc_create(&args);
 813         if (IS_ERR(client)) {
 814                 dprintk("NFSD: couldn't create callback client: %ld\n",
 815                         PTR_ERR(client));
 816                 return PTR_ERR(client);
 817         }
 818         cred = get_backchannel_cred(clp, client, ses);
 819         if (IS_ERR(cred)) {
 820                 rpc_shutdown_client(client);
 821                 return PTR_ERR(cred);
 822         }
 823         clp->cl_cb_client = client;
 824         clp->cl_cb_cred = cred;
 825         return 0;
 826 }
 827
 828 static void warn_no_callback_path(struct nfs4_client *clp, int reason)
 829 {
 830         dprintk("NFSD: warning: no callback path to client %.*s: error %d\n",
 831                 (int)clp->cl_name.len, clp->cl_name.data, reason);
 832 }
 833
 834 static void nfsd4_mark_cb_down(struct nfs4_client *clp, int reason)
 835 {
 836         if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
 837                 return;
 838         clp->cl_cb_state = NFSD4_CB_DOWN;
 839         warn_no_callback_path(clp, reason);
 840 }
 841
 842 static void nfsd4_mark_cb_fault(struct nfs4_client *clp, int reason)
 843 {
 844         if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
 845                 return;
 846         clp->cl_cb_state = NFSD4_CB_FAULT;
 847         warn_no_callback_path(clp, reason);
 848 }
 849
 850 static void nfsd4_cb_probe_done(struct rpc_task *task, void *calldata)
 851 {
 852         struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
 853
 854         if (task->tk_status)
 855                 nfsd4_mark_cb_down(clp, task->tk_status);
 856         else
 857                 clp->cl_cb_state = NFSD4_CB_UP;
 858 }
 859
 860 static const struct rpc_call_ops nfsd4_cb_probe_ops = {
 861         /* XXX: release method to ensure we set the cb channel down if
 862          * necessary on early failure? */
 863         .rpc_call_done = nfsd4_cb_probe_done,
 864 };
 865
 866 static struct workqueue_struct *callback_wq;
 867
 868 /*
 869  * Poke the callback thread to process any updates to the callback
 870  * parameters, and send a null probe.
 871  */
 872 void nfsd4_probe_callback(struct nfs4_client *clp)
 873 {
 874         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
 875         set_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
 876         nfsd4_run_cb(&clp->cl_cb_null);
 877 }
 878
 879 void nfsd4_probe_callback_sync(struct nfs4_client *clp)
 880 {
 881         nfsd4_probe_callback(clp);
 882         flush_workqueue(callback_wq);
 883 }
 884
 885 void nfsd4_change_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
 886 {
 887         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
 888         spin_lock(&clp->cl_lock);
 889         memcpy(&clp->cl_cb_conn, conn, sizeof(struct nfs4_cb_conn));
 890         spin_unlock(&clp->cl_lock);
 891 }
 892
 893 /*
 894  * There's currently a single callback channel slot.
 895  * If the slot is available, then mark it busy.  Otherwise, set the
 896  * thread for sleeping on the callback RPC wait queue.
 897  */
 898 static bool nfsd41_cb_get_slot(struct nfs4_client *clp, struct rpc_task *task)
 899 {
 900         if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
 901                 rpc_sleep_on(&clp->cl_cb_waitq, task, NULL);
 902                 /* Race breaker */
 903                 if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
 904                         dprintk("%s slot is busy\n", __func__);
 905                         return false;
 906                 }
 907                 rpc_wake_up_queued_task(&clp->cl_cb_waitq, task);
 908         }
 909         return true;
 910 }
 911
 912 /*
 913  * TODO: cb_sequence should support referring call lists, cachethis, multiple
 914  * slots, and mark callback channel down on communication errors.
 915  */
 916 static void nfsd4_cb_prepare(struct rpc_task *task, void *calldata)
 917 {
 918         struct nfsd4_callback *cb = calldata;
 919         struct nfs4_client *clp = cb->cb_clp;
 920         u32 minorversion = clp->cl_minorversion;
 921
 922         /*
 923          * cb_seq_status is only set in decode_cb_sequence4res,
 924          * and so will remain 1 if an rpc level failure occurs.
 925          */
 926         cb->cb_seq_status = 1;
 927         cb->cb_status = 0;
 928         if (minorversion) {
 929                 if (!nfsd41_cb_get_slot(clp, task))
 930                         return;
 931         }
 932         rpc_call_start(task);
 933 }
 934
 935 static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback *cb)
 936 {
 937         struct nfs4_client *clp = cb->cb_clp;
 938         struct nfsd4_session *session = clp->cl_cb_session;
 939         bool ret = true;
 940
 941         if (!clp->cl_minorversion) {
 942                 /*
 943                  * If the backchannel connection was shut down while this
 944                  * task was queued, we need to resubmit it after setting up
 945                  * a new backchannel connection.
 946                  *
 947                  * Note that if we lost our callback connection permanently
 948                  * the submission code will error out, so we don't need to
 949                  * handle that case here.
 950                  */
 951                 if (task->tk_flags & RPC_TASK_KILLED)
 952                         goto need_restart;
 953
 954                 return true;
 955         }
 956
 957         switch (cb->cb_seq_status) {
 958         case 0:
 959                 /*
 960                  * No need for lock, access serialized in nfsd4_cb_prepare
 961                  *
 962                  * RFC5661 20.9.3
 963                  * If CB_SEQUENCE returns an error, then the state of the slot
 964                  * (sequence ID, cached reply) MUST NOT change.
 965                  */
 966                 ++session->se_cb_seq_nr;
 967                 break;
 968         case -ESERVERFAULT:
 969                 ++session->se_cb_seq_nr;
 970         case 1:
 971         case -NFS4ERR_BADSESSION:
 972                 nfsd4_mark_cb_fault(cb->cb_clp, cb->cb_seq_status);
 973                 ret = false;
 974                 break;
 975         case -NFS4ERR_DELAY:
 976                 if (!rpc_restart_call(task))
 977                         goto out;
 978
 979                 rpc_delay(task, 2 * HZ);
 980                 return false;
 981         case -NFS4ERR_BADSLOT:
 982                 goto retry_nowait;
 983         case -NFS4ERR_SEQ_MISORDERED:
 984                 if (session->se_cb_seq_nr != 1) {
 985                         session->se_cb_seq_nr = 1;
 986                         goto retry_nowait;
 987                 }
 988                 break;
 989         default:
 990                 dprintk("%s: unprocessed error %d\n", __func__,
 991                         cb->cb_seq_status);
 992         }
 993
 994         clear_bit(0, &clp->cl_cb_slot_busy);
 995         rpc_wake_up_next(&clp->cl_cb_waitq);
 996         dprintk("%s: freed slot, new seqid=%d\n", __func__,
 997                 clp->cl_cb_session->se_cb_seq_nr);
 998
 999         if (task->tk_flags & RPC_TASK_KILLED)
1000                 goto need_restart;
1001 out:
1002         return ret;
1003 retry_nowait:
1004         if (rpc_restart_call_prepare(task))
1005                 ret = false;
1006         goto out;
1007 need_restart:
1008         task->tk_status = 0;
1009         cb->cb_need_restart = true;
1010         return false;
1011 }
1012
1013 static void nfsd4_cb_done(struct rpc_task *task, void *calldata)
1014 {
1015         struct nfsd4_callback *cb = calldata;
1016         struct nfs4_client *clp = cb->cb_clp;
1017
1018         dprintk("%s: minorversion=%d\n", __func__,
1019                 clp->cl_minorversion);
1020
1021         if (!nfsd4_cb_sequence_done(task, cb))
1022                 return;
1023
1024         if (cb->cb_status) {
1025                 WARN_ON_ONCE(task->tk_status);
1026                 task->tk_status = cb->cb_status;
1027         }
1028
1029         switch (cb->cb_ops->done(cb, task)) {
1030         case 0:
1031                 task->tk_status = 0;
1032                 rpc_restart_call_prepare(task);
1033                 return;
1034         case 1:
1035                 break;
1036         case -1:
1037                 /* Network partition? */
1038                 nfsd4_mark_cb_down(clp, task->tk_status);
1039                 break;
1040         default:
1041                 BUG();
1042         }
1043 }
1044
1045 static void nfsd4_cb_release(void *calldata)
1046 {
1047         struct nfsd4_callback *cb = calldata;
1048
1049         if (cb->cb_need_restart)
1050                 nfsd4_run_cb(cb);
1051         else
1052                 cb->cb_ops->release(cb);
1053
1054 }
1055
1056 static const struct rpc_call_ops nfsd4_cb_ops = {
1057         .rpc_call_prepare = nfsd4_cb_prepare,
1058         .rpc_call_done = nfsd4_cb_done,
1059         .rpc_release = nfsd4_cb_release,
1060 };
1061
1062 int nfsd4_create_callback_queue(void)
1063 {
1064         callback_wq = create_singlethread_workqueue("nfsd4_callbacks");
1065         if (!callback_wq)
1066                 return -ENOMEM;
1067         return 0;
1068 }
1069
1070 void nfsd4_destroy_callback_queue(void)
1071 {
1072         destroy_workqueue(callback_wq);
1073 }
1074
1075 /* must be called under the state lock */
1076 void nfsd4_shutdown_callback(struct nfs4_client *clp)
1077 {
1078         set_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags);
1079         /*
1080          * Note this won't actually result in a null callback;
1081          * instead, nfsd4_run_cb_null() will detect the killed
1082          * client, destroy the rpc client, and stop:
1083          */
1084         nfsd4_run_cb(&clp->cl_cb_null);
1085         flush_workqueue(callback_wq);
1086 }
1087
1088 /* requires cl_lock: */
1089 static struct nfsd4_conn * __nfsd4_find_backchannel(struct nfs4_client *clp)
1090 {
1091         struct nfsd4_session *s;
1092         struct nfsd4_conn *c;
1093
1094         list_for_each_entry(s, &clp->cl_sessions, se_perclnt) {
1095                 list_for_each_entry(c, &s->se_conns, cn_persession) {
1096                         if (c->cn_flags & NFS4_CDFC4_BACK)
1097                                 return c;
1098                 }
1099         }
1100         return NULL;
1101 }
1102
1103 static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
1104 {
1105         struct nfs4_cb_conn conn;
1106         struct nfs4_client *clp = cb->cb_clp;
1107         struct nfsd4_session *ses = NULL;
1108         struct nfsd4_conn *c;
1109         int err;
1110
1111         /*
1112          * This is either an update, or the client dying; in either case,
1113          * kill the old client:
1114          */
1115         if (clp->cl_cb_client) {
1116                 rpc_shutdown_client(clp->cl_cb_client);
1117                 clp->cl_cb_client = NULL;
1118                 put_rpccred(clp->cl_cb_cred);
1119                 clp->cl_cb_cred = NULL;
1120         }
1121         if (clp->cl_cb_conn.cb_xprt) {
1122                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1123                 clp->cl_cb_conn.cb_xprt = NULL;
1124         }
1125         if (test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags))
1126                 return;
1127         spin_lock(&clp->cl_lock);
1128         /*
1129          * Only serialized callback code is allowed to clear these
1130          * flags; main nfsd code can only set them:
1131          */
1132         BUG_ON(!(clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK));
1133         clear_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
1134         memcpy(&conn, &cb->cb_clp->cl_cb_conn, sizeof(struct nfs4_cb_conn));
1135         c = __nfsd4_find_backchannel(clp);
1136         if (c) {
1137                 svc_xprt_get(c->cn_xprt);
1138                 conn.cb_xprt = c->cn_xprt;
1139                 ses = c->cn_session;
1140         }
1141         spin_unlock(&clp->cl_lock);
1142
1143         err = setup_callback_client(clp, &conn, ses);
1144         if (err) {
1145                 nfsd4_mark_cb_down(clp, err);
1146                 return;
1147         }
1148 }
1149
1150 static void
1151 nfsd4_run_cb_work(struct work_struct *work)
1152 {
1153         struct nfsd4_callback *cb =
1154                 container_of(work, struct nfsd4_callback, cb_work);
1155         struct nfs4_client *clp = cb->cb_clp;
1156         struct rpc_clnt *clnt;
1157
1158         if (cb->cb_need_restart) {
1159                 cb->cb_need_restart = false;
1160         } else {
1161                 if (cb->cb_ops && cb->cb_ops->prepare)
1162                         cb->cb_ops->prepare(cb);
1163         }
1164
1165         if (clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK)
1166                 nfsd4_process_cb_update(cb);
1167
1168         clnt = clp->cl_cb_client;
1169         if (!clnt) {
1170                 /* Callback channel broken, or client killed; give up: */
1171                 if (cb->cb_ops && cb->cb_ops->release)
1172                         cb->cb_ops->release(cb);
1173                 return;
1174         }
1175
1176         /*
1177          * Don't send probe messages for 4.1 or later.
1178          */
1179         if (!cb->cb_ops && clp->cl_minorversion) {
1180                 clp->cl_cb_state = NFSD4_CB_UP;
1181                 return;
1182         }
1183
1184         cb->cb_msg.rpc_cred = clp->cl_cb_cred;
1185         rpc_call_async(clnt, &cb->cb_msg, RPC_TASK_SOFT | RPC_TASK_SOFTCONN,
1186                         cb->cb_ops ? &nfsd4_cb_ops : &nfsd4_cb_probe_ops, cb);
1187 }
1188
1189 void nfsd4_init_cb(struct nfsd4_callback *cb, struct nfs4_client *clp,
1190                 const struct nfsd4_callback_ops *ops, enum nfsd4_cb_op op)
1191 {
1192         cb->cb_clp = clp;
1193         cb->cb_msg.rpc_proc = &nfs4_cb_procedures[op];
1194         cb->cb_msg.rpc_argp = cb;
1195         cb->cb_msg.rpc_resp = cb;
1196         cb->cb_ops = ops;
1197         INIT_WORK(&cb->cb_work, nfsd4_run_cb_work);
1198         cb->cb_seq_status = 1;
1199         cb->cb_status = 0;
1200         cb->cb_need_restart = false;
1201 }
1202
1203 void nfsd4_run_cb(struct nfsd4_callback *cb)
1204 {
1205         queue_work(callback_wq, &cb->cb_work);
1206 }