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         p += XDR_QUADLEN(length);
 307         hdr->nops = be32_to_cpup(p);
 308         return 0;
 309 out_overflow:
 310         print_overflow_msg(__func__, xdr);
 311         return -EIO;
 312 }
 313
 314 /*
 315  * CB_RECALL4args
 316  *
 317  *      struct CB_RECALL4args {
 318  *              stateid4        stateid;
 319  *              bool            truncate;
 320  *              nfs_fh4         fh;
 321  *      };
 322  */
 323 static void encode_cb_recall4args(struct xdr_stream *xdr,
 324                                   const struct nfs4_delegation *dp,
 325                                   struct nfs4_cb_compound_hdr *hdr)
 326 {
 327         __be32 *p;
 328
 329         encode_nfs_cb_opnum4(xdr, OP_CB_RECALL);
 330         encode_stateid4(xdr, &dp->dl_stid.sc_stateid);
 331
 332         p = xdr_reserve_space(xdr, 4);
 333         *p++ = xdr_zero;                        /* truncate */
 334
 335         encode_nfs_fh4(xdr, &dp->dl_stid.sc_file->fi_fhandle);
 336
 337         hdr->nops++;
 338 }
 339
 340 /*
 341  * CB_SEQUENCE4args
 342  *
 343  *      struct CB_SEQUENCE4args {
 344  *              sessionid4              csa_sessionid;
 345  *              sequenceid4             csa_sequenceid;
 346  *              slotid4                 csa_slotid;
 347  *              slotid4                 csa_highest_slotid;
 348  *              bool                    csa_cachethis;
 349  *              referring_call_list4    csa_referring_call_lists<>;
 350  *      };
 351  */
 352 static void encode_cb_sequence4args(struct xdr_stream *xdr,
 353                                     const struct nfsd4_callback *cb,
 354                                     struct nfs4_cb_compound_hdr *hdr)
 355 {
 356         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
 357         __be32 *p;
 358
 359         if (hdr->minorversion == 0)
 360                 return;
 361
 362         encode_nfs_cb_opnum4(xdr, OP_CB_SEQUENCE);
 363         encode_sessionid4(xdr, session);
 364
 365         p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4 + 4);
 366         *p++ = cpu_to_be32(session->se_cb_seq_nr);      /* csa_sequenceid */
 367         *p++ = xdr_zero;                        /* csa_slotid */
 368         *p++ = xdr_zero;                        /* csa_highest_slotid */
 369         *p++ = xdr_zero;                        /* csa_cachethis */
 370         xdr_encode_empty_array(p);              /* csa_referring_call_lists */
 371
 372         hdr->nops++;
 373 }
 374
 375 /*
 376  * CB_SEQUENCE4resok
 377  *
 378  *      struct CB_SEQUENCE4resok {
 379  *              sessionid4      csr_sessionid;
 380  *              sequenceid4     csr_sequenceid;
 381  *              slotid4         csr_slotid;
 382  *              slotid4         csr_highest_slotid;
 383  *              slotid4         csr_target_highest_slotid;
 384  *      };
 385  *
 386  *      union CB_SEQUENCE4res switch (nfsstat4 csr_status) {
 387  *      case NFS4_OK:
 388  *              CB_SEQUENCE4resok       csr_resok4;
 389  *      default:
 390  *              void;
 391  *      };
 392  *
 393  * Our current back channel implmentation supports a single backchannel
 394  * with a single slot.
 395  */
 396 static int decode_cb_sequence4resok(struct xdr_stream *xdr,
 397                                     struct nfsd4_callback *cb)
 398 {
 399         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
 400         int status = -ESERVERFAULT;
 401         __be32 *p;
 402         u32 dummy;
 403
 404         /*
 405          * If the server returns different values for sessionID, slotID or
 406          * sequence number, the server is looney tunes.
 407          */
 408         p = xdr_inline_decode(xdr, NFS4_MAX_SESSIONID_LEN + 4 + 4 + 4 + 4);
 409         if (unlikely(p == NULL))
 410                 goto out_overflow;
 411
 412         if (memcmp(p, session->se_sessionid.data, NFS4_MAX_SESSIONID_LEN)) {
 413                 dprintk("NFS: %s Invalid session id\n", __func__);
 414                 goto out;
 415         }
 416         p += XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN);
 417
 418         dummy = be32_to_cpup(p++);
 419         if (dummy != session->se_cb_seq_nr) {
 420                 dprintk("NFS: %s Invalid sequence number\n", __func__);
 421                 goto out;
 422         }
 423
 424         dummy = be32_to_cpup(p++);
 425         if (dummy != 0) {
 426                 dprintk("NFS: %s Invalid slotid\n", __func__);
 427                 goto out;
 428         }
 429
 430         /*
 431          * FIXME: process highest slotid and target highest slotid
 432          */
 433         status = 0;
 434 out:
 435         cb->cb_seq_status = status;
 436         return status;
 437 out_overflow:
 438         print_overflow_msg(__func__, xdr);
 439         status = -EIO;
 440         goto out;
 441 }
 442
 443 static int decode_cb_sequence4res(struct xdr_stream *xdr,
 444                                   struct nfsd4_callback *cb)
 445 {
 446         int status;
 447
 448         if (cb->cb_clp->cl_minorversion == 0)
 449                 return 0;
 450
 451         status = decode_cb_op_status(xdr, OP_CB_SEQUENCE, &cb->cb_seq_status);
 452         if (unlikely(status || cb->cb_seq_status))
 453                 return status;
 454
 455         return decode_cb_sequence4resok(xdr, cb);
 456 }
 457
 458 /*
 459  * NFSv4.0 and NFSv4.1 XDR encode functions
 460  *
 461  * NFSv4.0 callback argument types are defined in section 15 of RFC
 462  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
 463  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
 464  * Protocol".
 465  */
 466
 467 /*
 468  * NB: Without this zero space reservation, callbacks over krb5p fail
 469  */
 470 static void nfs4_xdr_enc_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
 471                                  void *__unused)
 472 {
 473         xdr_reserve_space(xdr, 0);
 474 }
 475
 476 /*
 477  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
 478  */
 479 static void nfs4_xdr_enc_cb_recall(struct rpc_rqst *req, struct xdr_stream *xdr,
 480                                    const struct nfsd4_callback *cb)
 481 {
 482         const struct nfs4_delegation *dp = cb_to_delegation(cb);
 483         struct nfs4_cb_compound_hdr hdr = {
 484                 .ident = cb->cb_clp->cl_cb_ident,
 485                 .minorversion = cb->cb_clp->cl_minorversion,
 486         };
 487
 488         encode_cb_compound4args(xdr, &hdr);
 489         encode_cb_sequence4args(xdr, cb, &hdr);
 490         encode_cb_recall4args(xdr, dp, &hdr);
 491         encode_cb_nops(&hdr);
 492 }
 493
 494
 495 /*
 496  * NFSv4.0 and NFSv4.1 XDR decode functions
 497  *
 498  * NFSv4.0 callback result types are defined in section 15 of RFC
 499  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
 500  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
 501  * Protocol".
 502  */
 503
 504 static int nfs4_xdr_dec_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
 505                                 void *__unused)
 506 {
 507         return 0;
 508 }
 509
 510 /*
 511  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
 512  */
 513 static int nfs4_xdr_dec_cb_recall(struct rpc_rqst *rqstp,
 514                                   struct xdr_stream *xdr,
 515                                   struct nfsd4_callback *cb)
 516 {
 517         struct nfs4_cb_compound_hdr hdr;
 518         int status;
 519
 520         status = decode_cb_compound4res(xdr, &hdr);
 521         if (unlikely(status))
 522                 return status;
 523
 524         if (cb != NULL) {
 525                 status = decode_cb_sequence4res(xdr, cb);
 526                 if (unlikely(status || cb->cb_seq_status))
 527                         return status;
 528         }
 529
 530         return decode_cb_op_status(xdr, OP_CB_RECALL, &cb->cb_status);
 531 }
 532
 533 #ifdef CONFIG_NFSD_PNFS
 534 /*
 535  * CB_LAYOUTRECALL4args
 536  *
 537  *      struct layoutrecall_file4 {
 538  *              nfs_fh4         lor_fh;
 539  *              offset4         lor_offset;
 540  *              length4         lor_length;
 541  *              stateid4        lor_stateid;
 542  *      };
 543  *
 544  *      union layoutrecall4 switch(layoutrecall_type4 lor_recalltype) {
 545  *      case LAYOUTRECALL4_FILE:
 546  *              layoutrecall_file4 lor_layout;
 547  *      case LAYOUTRECALL4_FSID:
 548  *              fsid4              lor_fsid;
 549  *      case LAYOUTRECALL4_ALL:
 550  *              void;
 551  *      };
 552  *
 553  *      struct CB_LAYOUTRECALL4args {
 554  *              layouttype4             clora_type;
 555  *              layoutiomode4           clora_iomode;
 556  *              bool                    clora_changed;
 557  *              layoutrecall4           clora_recall;
 558  *      };
 559  */
 560 static void encode_cb_layout4args(struct xdr_stream *xdr,
 561                                   const struct nfs4_layout_stateid *ls,
 562                                   struct nfs4_cb_compound_hdr *hdr)
 563 {
 564         __be32 *p;
 565
 566         BUG_ON(hdr->minorversion == 0);
 567
 568         p = xdr_reserve_space(xdr, 5 * 4);
 569         *p++ = cpu_to_be32(OP_CB_LAYOUTRECALL);
 570         *p++ = cpu_to_be32(ls->ls_layout_type);
 571         *p++ = cpu_to_be32(IOMODE_ANY);
 572         *p++ = cpu_to_be32(1);
 573         *p = cpu_to_be32(RETURN_FILE);
 574
 575         encode_nfs_fh4(xdr, &ls->ls_stid.sc_file->fi_fhandle);
 576
 577         p = xdr_reserve_space(xdr, 2 * 8);
 578         p = xdr_encode_hyper(p, 0);
 579         xdr_encode_hyper(p, NFS4_MAX_UINT64);
 580
 581         encode_stateid4(xdr, &ls->ls_recall_sid);
 582
 583         hdr->nops++;
 584 }
 585
 586 static void nfs4_xdr_enc_cb_layout(struct rpc_rqst *req,
 587                                    struct xdr_stream *xdr,
 588                                    const struct nfsd4_callback *cb)
 589 {
 590         const struct nfs4_layout_stateid *ls =
 591                 container_of(cb, struct nfs4_layout_stateid, ls_recall);
 592         struct nfs4_cb_compound_hdr hdr = {
 593                 .ident = 0,
 594                 .minorversion = cb->cb_clp->cl_minorversion,
 595         };
 596
 597         encode_cb_compound4args(xdr, &hdr);
 598         encode_cb_sequence4args(xdr, cb, &hdr);
 599         encode_cb_layout4args(xdr, ls, &hdr);
 600         encode_cb_nops(&hdr);
 601 }
 602
 603 static int nfs4_xdr_dec_cb_layout(struct rpc_rqst *rqstp,
 604                                   struct xdr_stream *xdr,
 605                                   struct nfsd4_callback *cb)
 606 {
 607         struct nfs4_cb_compound_hdr hdr;
 608         int status;
 609
 610         status = decode_cb_compound4res(xdr, &hdr);
 611         if (unlikely(status))
 612                 return status;
 613
 614         if (cb) {
 615                 status = decode_cb_sequence4res(xdr, cb);
 616                 if (unlikely(status || cb->cb_seq_status))
 617                         return status;
 618         }
 619         return decode_cb_op_status(xdr, OP_CB_LAYOUTRECALL, &cb->cb_status);
 620 }
 621 #endif /* CONFIG_NFSD_PNFS */
 622
 623 static void encode_stateowner(struct xdr_stream *xdr, struct nfs4_stateowner *so)
 624 {
 625         __be32  *p;
 626
 627         p = xdr_reserve_space(xdr, 8 + 4 + so->so_owner.len);
 628         p = xdr_encode_opaque_fixed(p, &so->so_client->cl_clientid, 8);
 629         xdr_encode_opaque(p, so->so_owner.data, so->so_owner.len);
 630 }
 631
 632 static void nfs4_xdr_enc_cb_notify_lock(struct rpc_rqst *req,
 633                                         struct xdr_stream *xdr,
 634                                         const struct nfsd4_callback *cb)
 635 {
 636         const struct nfsd4_blocked_lock *nbl =
 637                 container_of(cb, struct nfsd4_blocked_lock, nbl_cb);
 638         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)nbl->nbl_lock.fl_owner;
 639         struct nfs4_cb_compound_hdr hdr = {
 640                 .ident = 0,
 641                 .minorversion = cb->cb_clp->cl_minorversion,
 642         };
 643
 644         __be32 *p;
 645
 646         BUG_ON(hdr.minorversion == 0);
 647
 648         encode_cb_compound4args(xdr, &hdr);
 649         encode_cb_sequence4args(xdr, cb, &hdr);
 650
 651         p = xdr_reserve_space(xdr, 4);
 652         *p = cpu_to_be32(OP_CB_NOTIFY_LOCK);
 653         encode_nfs_fh4(xdr, &nbl->nbl_fh);
 654         encode_stateowner(xdr, &lo->lo_owner);
 655         hdr.nops++;
 656
 657         encode_cb_nops(&hdr);
 658 }
 659
 660 static int nfs4_xdr_dec_cb_notify_lock(struct rpc_rqst *rqstp,
 661                                         struct xdr_stream *xdr,
 662                                         struct nfsd4_callback *cb)
 663 {
 664         struct nfs4_cb_compound_hdr hdr;
 665         int status;
 666
 667         status = decode_cb_compound4res(xdr, &hdr);
 668         if (unlikely(status))
 669                 return status;
 670
 671         if (cb) {
 672                 status = decode_cb_sequence4res(xdr, cb);
 673                 if (unlikely(status || cb->cb_seq_status))
 674                         return status;
 675         }
 676         return decode_cb_op_status(xdr, OP_CB_NOTIFY_LOCK, &cb->cb_status);
 677 }
 678
 679 /*
 680  * RPC procedure tables
 681  */
 682 #define PROC(proc, call, argtype, restype)                              \
 683 [NFSPROC4_CLNT_##proc] = {                                              \
 684         .p_proc    = NFSPROC4_CB_##call,                                \
 685         .p_encode  = (kxdreproc_t)nfs4_xdr_enc_##argtype,               \
 686         .p_decode  = (kxdrdproc_t)nfs4_xdr_dec_##restype,               \
 687         .p_arglen  = NFS4_enc_##argtype##_sz,                           \
 688         .p_replen  = NFS4_dec_##restype##_sz,                           \
 689         .p_statidx = NFSPROC4_CB_##call,                                \
 690         .p_name    = #proc,                                             \
 691 }
 692
 693 static struct rpc_procinfo nfs4_cb_procedures[] = {
 694         PROC(CB_NULL,   NULL,           cb_null,        cb_null),
 695         PROC(CB_RECALL, COMPOUND,       cb_recall,      cb_recall),
 696 #ifdef CONFIG_NFSD_PNFS
 697         PROC(CB_LAYOUT, COMPOUND,       cb_layout,      cb_layout),
 698 #endif
 699         PROC(CB_NOTIFY_LOCK,    COMPOUND,       cb_notify_lock, cb_notify_lock),
 700 };
 701
 702 static struct rpc_version nfs_cb_version4 = {
 703 /*
 704  * Note on the callback rpc program version number: despite language in rfc
 705  * 5661 section 18.36.3 requiring servers to use 4 in this field, the
 706  * official xdr descriptions for both 4.0 and 4.1 specify version 1, and
 707  * in practice that appears to be what implementations use.  The section
 708  * 18.36.3 language is expected to be fixed in an erratum.
 709  */
 710         .number                 = 1,
 711         .nrprocs                = ARRAY_SIZE(nfs4_cb_procedures),
 712         .procs                  = nfs4_cb_procedures
 713 };
 714
 715 static const struct rpc_version *nfs_cb_version[] = {
 716         &nfs_cb_version4,
 717 };
 718
 719 static const struct rpc_program cb_program;
 720
 721 static struct rpc_stat cb_stats = {
 722         .program                = &cb_program
 723 };
 724
 725 #define NFS4_CALLBACK 0x40000000
 726 static const struct rpc_program cb_program = {
 727         .name                   = "nfs4_cb",
 728         .number                 = NFS4_CALLBACK,
 729         .nrvers                 = ARRAY_SIZE(nfs_cb_version),
 730         .version                = nfs_cb_version,
 731         .stats                  = &cb_stats,
 732         .pipe_dir_name          = "nfsd4_cb",
 733 };
 734
 735 static int max_cb_time(struct net *net)
 736 {
 737         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 738         return max(nn->nfsd4_lease/10, (time_t)1) * HZ;
 739 }
 740
 741 static struct rpc_cred *callback_cred;
 742
 743 int set_callback_cred(void)
 744 {
 745         if (callback_cred)
 746                 return 0;
 747         callback_cred = rpc_lookup_machine_cred("nfs");
 748         if (!callback_cred)
 749                 return -ENOMEM;
 750         return 0;
 751 }
 752
 753 void cleanup_callback_cred(void)
 754 {
 755         if (callback_cred) {
 756                 put_rpccred(callback_cred);
 757                 callback_cred = NULL;
 758         }
 759 }
 760
 761 static struct rpc_cred *get_backchannel_cred(struct nfs4_client *clp, struct rpc_clnt *client, struct nfsd4_session *ses)
 762 {
 763         if (clp->cl_minorversion == 0) {
 764                 return get_rpccred(callback_cred);
 765         } else {
 766                 struct rpc_auth *auth = client->cl_auth;
 767                 struct auth_cred acred = {};
 768
 769                 acred.uid = ses->se_cb_sec.uid;
 770                 acred.gid = ses->se_cb_sec.gid;
 771                 return auth->au_ops->lookup_cred(client->cl_auth, &acred, 0);
 772         }
 773 }
 774
 775 static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn, struct nfsd4_session *ses)
 776 {
 777         int maxtime = max_cb_time(clp->net);
 778         struct rpc_timeout      timeparms = {
 779                 .to_initval     = maxtime,
 780                 .to_retries     = 0,
 781                 .to_maxval      = maxtime,
 782         };
 783         struct rpc_create_args args = {
 784                 .net            = clp->net,
 785                 .address        = (struct sockaddr *) &conn->cb_addr,
 786                 .addrsize       = conn->cb_addrlen,
 787                 .saddress       = (struct sockaddr *) &conn->cb_saddr,
 788                 .timeout        = &timeparms,
 789                 .program        = &cb_program,
 790                 .version        = 0,
 791                 .flags          = (RPC_CLNT_CREATE_NOPING | RPC_CLNT_CREATE_QUIET),
 792         };
 793         struct rpc_clnt *client;
 794         struct rpc_cred *cred;
 795
 796         if (clp->cl_minorversion == 0) {
 797                 if (!clp->cl_cred.cr_principal &&
 798                                 (clp->cl_cred.cr_flavor >= RPC_AUTH_GSS_KRB5))
 799                         return -EINVAL;
 800                 args.client_name = clp->cl_cred.cr_principal;
 801                 args.prognumber = conn->cb_prog;
 802                 args.protocol = XPRT_TRANSPORT_TCP;
 803                 args.authflavor = clp->cl_cred.cr_flavor;
 804                 clp->cl_cb_ident = conn->cb_ident;
 805         } else {
 806                 if (!conn->cb_xprt)
 807                         return -EINVAL;
 808                 clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
 809                 clp->cl_cb_session = ses;
 810                 args.bc_xprt = conn->cb_xprt;
 811                 args.prognumber = clp->cl_cb_session->se_cb_prog;
 812                 args.protocol = conn->cb_xprt->xpt_class->xcl_ident |
 813                                 XPRT_TRANSPORT_BC;
 814                 args.authflavor = ses->se_cb_sec.flavor;
 815         }
 816         /* Create RPC client */
 817         client = rpc_create(&args);
 818         if (IS_ERR(client)) {
 819                 dprintk("NFSD: couldn't create callback client: %ld\n",
 820                         PTR_ERR(client));
 821                 return PTR_ERR(client);
 822         }
 823         cred = get_backchannel_cred(clp, client, ses);
 824         if (IS_ERR(cred)) {
 825                 rpc_shutdown_client(client);
 826                 return PTR_ERR(cred);
 827         }
 828         clp->cl_cb_client = client;
 829         clp->cl_cb_cred = cred;
 830         return 0;
 831 }
 832
 833 static void warn_no_callback_path(struct nfs4_client *clp, int reason)
 834 {
 835         dprintk("NFSD: warning: no callback path to client %.*s: error %d\n",
 836                 (int)clp->cl_name.len, clp->cl_name.data, reason);
 837 }
 838
 839 static void nfsd4_mark_cb_down(struct nfs4_client *clp, int reason)
 840 {
 841         if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
 842                 return;
 843         clp->cl_cb_state = NFSD4_CB_DOWN;
 844         warn_no_callback_path(clp, reason);
 845 }
 846
 847 static void nfsd4_mark_cb_fault(struct nfs4_client *clp, int reason)
 848 {
 849         if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
 850                 return;
 851         clp->cl_cb_state = NFSD4_CB_FAULT;
 852         warn_no_callback_path(clp, reason);
 853 }
 854
 855 static void nfsd4_cb_probe_done(struct rpc_task *task, void *calldata)
 856 {
 857         struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
 858
 859         if (task->tk_status)
 860                 nfsd4_mark_cb_down(clp, task->tk_status);
 861         else
 862                 clp->cl_cb_state = NFSD4_CB_UP;
 863 }
 864
 865 static const struct rpc_call_ops nfsd4_cb_probe_ops = {
 866         /* XXX: release method to ensure we set the cb channel down if
 867          * necessary on early failure? */
 868         .rpc_call_done = nfsd4_cb_probe_done,
 869 };
 870
 871 static struct workqueue_struct *callback_wq;
 872
 873 /*
 874  * Poke the callback thread to process any updates to the callback
 875  * parameters, and send a null probe.
 876  */
 877 void nfsd4_probe_callback(struct nfs4_client *clp)
 878 {
 879         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
 880         set_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
 881         nfsd4_run_cb(&clp->cl_cb_null);
 882 }
 883
 884 void nfsd4_probe_callback_sync(struct nfs4_client *clp)
 885 {
 886         nfsd4_probe_callback(clp);
 887         flush_workqueue(callback_wq);
 888 }
 889
 890 void nfsd4_change_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
 891 {
 892         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
 893         spin_lock(&clp->cl_lock);
 894         memcpy(&clp->cl_cb_conn, conn, sizeof(struct nfs4_cb_conn));
 895         spin_unlock(&clp->cl_lock);
 896 }
 897
 898 /*
 899  * There's currently a single callback channel slot.
 900  * If the slot is available, then mark it busy.  Otherwise, set the
 901  * thread for sleeping on the callback RPC wait queue.
 902  */
 903 static bool nfsd41_cb_get_slot(struct nfs4_client *clp, struct rpc_task *task)
 904 {
 905         if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
 906                 rpc_sleep_on(&clp->cl_cb_waitq, task, NULL);
 907                 /* Race breaker */
 908                 if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
 909                         dprintk("%s slot is busy\n", __func__);
 910                         return false;
 911                 }
 912                 rpc_wake_up_queued_task(&clp->cl_cb_waitq, task);
 913         }
 914         return true;
 915 }
 916
 917 /*
 918  * TODO: cb_sequence should support referring call lists, cachethis, multiple
 919  * slots, and mark callback channel down on communication errors.
 920  */
 921 static void nfsd4_cb_prepare(struct rpc_task *task, void *calldata)
 922 {
 923         struct nfsd4_callback *cb = calldata;
 924         struct nfs4_client *clp = cb->cb_clp;
 925         u32 minorversion = clp->cl_minorversion;
 926
 927         /*
 928          * cb_seq_status is only set in decode_cb_sequence4res,
 929          * and so will remain 1 if an rpc level failure occurs.
 930          */
 931         cb->cb_seq_status = 1;
 932         cb->cb_status = 0;
 933         if (minorversion) {
 934                 if (!nfsd41_cb_get_slot(clp, task))
 935                         return;
 936         }
 937         rpc_call_start(task);
 938 }
 939
 940 static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback *cb)
 941 {
 942         struct nfs4_client *clp = cb->cb_clp;
 943         struct nfsd4_session *session = clp->cl_cb_session;
 944         bool ret = true;
 945
 946         if (!clp->cl_minorversion) {
 947                 /*
 948                  * If the backchannel connection was shut down while this
 949                  * task was queued, we need to resubmit it after setting up
 950                  * a new backchannel connection.
 951                  *
 952                  * Note that if we lost our callback connection permanently
 953                  * the submission code will error out, so we don't need to
 954                  * handle that case here.
 955                  */
 956                 if (task->tk_flags & RPC_TASK_KILLED)
 957                         goto need_restart;
 958
 959                 return true;
 960         }
 961
 962         switch (cb->cb_seq_status) {
 963         case 0:
 964                 /*
 965                  * No need for lock, access serialized in nfsd4_cb_prepare
 966                  *
 967                  * RFC5661 20.9.3
 968                  * If CB_SEQUENCE returns an error, then the state of the slot
 969                  * (sequence ID, cached reply) MUST NOT change.
 970                  */
 971                 ++session->se_cb_seq_nr;
 972                 break;
 973         case -ESERVERFAULT:
 974                 ++session->se_cb_seq_nr;
 975         case 1:
 976         case -NFS4ERR_BADSESSION:
 977                 nfsd4_mark_cb_fault(cb->cb_clp, cb->cb_seq_status);
 978                 ret = false;
 979                 break;
 980         case -NFS4ERR_DELAY:
 981                 if (!rpc_restart_call(task))
 982                         goto out;
 983
 984                 rpc_delay(task, 2 * HZ);
 985                 return false;
 986         case -NFS4ERR_BADSLOT:
 987                 goto retry_nowait;
 988         case -NFS4ERR_SEQ_MISORDERED:
 989                 if (session->se_cb_seq_nr != 1) {
 990                         session->se_cb_seq_nr = 1;
 991                         goto retry_nowait;
 992                 }
 993                 break;
 994         default:
 995                 dprintk("%s: unprocessed error %d\n", __func__,
 996                         cb->cb_seq_status);
 997         }
 998
 999         clear_bit(0, &clp->cl_cb_slot_busy);
1000         rpc_wake_up_next(&clp->cl_cb_waitq);
1001         dprintk("%s: freed slot, new seqid=%d\n", __func__,
1002                 clp->cl_cb_session->se_cb_seq_nr);
1003
1004         if (task->tk_flags & RPC_TASK_KILLED)
1005                 goto need_restart;
1006 out:
1007         return ret;
1008 retry_nowait:
1009         if (rpc_restart_call_prepare(task))
1010                 ret = false;
1011         goto out;
1012 need_restart:
1013         task->tk_status = 0;
1014         cb->cb_need_restart = true;
1015         return false;
1016 }
1017
1018 static void nfsd4_cb_done(struct rpc_task *task, void *calldata)
1019 {
1020         struct nfsd4_callback *cb = calldata;
1021         struct nfs4_client *clp = cb->cb_clp;
1022
1023         dprintk("%s: minorversion=%d\n", __func__,
1024                 clp->cl_minorversion);
1025
1026         if (!nfsd4_cb_sequence_done(task, cb))
1027                 return;
1028
1029         if (cb->cb_status) {
1030                 WARN_ON_ONCE(task->tk_status);
1031                 task->tk_status = cb->cb_status;
1032         }
1033
1034         switch (cb->cb_ops->done(cb, task)) {
1035         case 0:
1036                 task->tk_status = 0;
1037                 rpc_restart_call_prepare(task);
1038                 return;
1039         case 1:
1040                 break;
1041         case -1:
1042                 /* Network partition? */
1043                 nfsd4_mark_cb_down(clp, task->tk_status);
1044                 break;
1045         default:
1046                 BUG();
1047         }
1048 }
1049
1050 static void nfsd4_cb_release(void *calldata)
1051 {
1052         struct nfsd4_callback *cb = calldata;
1053
1054         if (cb->cb_need_restart)
1055                 nfsd4_run_cb(cb);
1056         else
1057                 cb->cb_ops->release(cb);
1058
1059 }
1060
1061 static const struct rpc_call_ops nfsd4_cb_ops = {
1062         .rpc_call_prepare = nfsd4_cb_prepare,
1063         .rpc_call_done = nfsd4_cb_done,
1064         .rpc_release = nfsd4_cb_release,
1065 };
1066
1067 int nfsd4_create_callback_queue(void)
1068 {
1069         callback_wq = alloc_ordered_workqueue("nfsd4_callbacks", 0);
1070         if (!callback_wq)
1071                 return -ENOMEM;
1072         return 0;
1073 }
1074
1075 void nfsd4_destroy_callback_queue(void)
1076 {
1077         destroy_workqueue(callback_wq);
1078 }
1079
1080 /* must be called under the state lock */
1081 void nfsd4_shutdown_callback(struct nfs4_client *clp)
1082 {
1083         set_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags);
1084         /*
1085          * Note this won't actually result in a null callback;
1086          * instead, nfsd4_run_cb_null() will detect the killed
1087          * client, destroy the rpc client, and stop:
1088          */
1089         nfsd4_run_cb(&clp->cl_cb_null);
1090         flush_workqueue(callback_wq);
1091 }
1092
1093 /* requires cl_lock: */
1094 static struct nfsd4_conn * __nfsd4_find_backchannel(struct nfs4_client *clp)
1095 {
1096         struct nfsd4_session *s;
1097         struct nfsd4_conn *c;
1098
1099         list_for_each_entry(s, &clp->cl_sessions, se_perclnt) {
1100                 list_for_each_entry(c, &s->se_conns, cn_persession) {
1101                         if (c->cn_flags & NFS4_CDFC4_BACK)
1102                                 return c;
1103                 }
1104         }
1105         return NULL;
1106 }
1107
1108 static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
1109 {
1110         struct nfs4_cb_conn conn;
1111         struct nfs4_client *clp = cb->cb_clp;
1112         struct nfsd4_session *ses = NULL;
1113         struct nfsd4_conn *c;
1114         int err;
1115
1116         /*
1117          * This is either an update, or the client dying; in either case,
1118          * kill the old client:
1119          */
1120         if (clp->cl_cb_client) {
1121                 rpc_shutdown_client(clp->cl_cb_client);
1122                 clp->cl_cb_client = NULL;
1123                 put_rpccred(clp->cl_cb_cred);
1124                 clp->cl_cb_cred = NULL;
1125         }
1126         if (clp->cl_cb_conn.cb_xprt) {
1127                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1128                 clp->cl_cb_conn.cb_xprt = NULL;
1129         }
1130         if (test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags))
1131                 return;
1132         spin_lock(&clp->cl_lock);
1133         /*
1134          * Only serialized callback code is allowed to clear these
1135          * flags; main nfsd code can only set them:
1136          */
1137         BUG_ON(!(clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK));
1138         clear_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
1139         memcpy(&conn, &cb->cb_clp->cl_cb_conn, sizeof(struct nfs4_cb_conn));
1140         c = __nfsd4_find_backchannel(clp);
1141         if (c) {
1142                 svc_xprt_get(c->cn_xprt);
1143                 conn.cb_xprt = c->cn_xprt;
1144                 ses = c->cn_session;
1145         }
1146         spin_unlock(&clp->cl_lock);
1147
1148         err = setup_callback_client(clp, &conn, ses);
1149         if (err) {
1150                 nfsd4_mark_cb_down(clp, err);
1151                 return;
1152         }
1153 }
1154
1155 static void
1156 nfsd4_run_cb_work(struct work_struct *work)
1157 {
1158         struct nfsd4_callback *cb =
1159                 container_of(work, struct nfsd4_callback, cb_work);
1160         struct nfs4_client *clp = cb->cb_clp;
1161         struct rpc_clnt *clnt;
1162
1163         if (cb->cb_need_restart) {
1164                 cb->cb_need_restart = false;
1165         } else {
1166                 if (cb->cb_ops && cb->cb_ops->prepare)
1167                         cb->cb_ops->prepare(cb);
1168         }
1169
1170         if (clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK)
1171                 nfsd4_process_cb_update(cb);
1172
1173         clnt = clp->cl_cb_client;
1174         if (!clnt) {
1175                 /* Callback channel broken, or client killed; give up: */
1176                 if (cb->cb_ops && cb->cb_ops->release)
1177                         cb->cb_ops->release(cb);
1178                 return;
1179         }
1180
1181         /*
1182          * Don't send probe messages for 4.1 or later.
1183          */
1184         if (!cb->cb_ops && clp->cl_minorversion) {
1185                 clp->cl_cb_state = NFSD4_CB_UP;
1186                 return;
1187         }
1188
1189         cb->cb_msg.rpc_cred = clp->cl_cb_cred;
1190         rpc_call_async(clnt, &cb->cb_msg, RPC_TASK_SOFT | RPC_TASK_SOFTCONN,
1191                         cb->cb_ops ? &nfsd4_cb_ops : &nfsd4_cb_probe_ops, cb);
1192 }
1193
1194 void nfsd4_init_cb(struct nfsd4_callback *cb, struct nfs4_client *clp,
1195                 const struct nfsd4_callback_ops *ops, enum nfsd4_cb_op op)
1196 {
1197         cb->cb_clp = clp;
1198         cb->cb_msg.rpc_proc = &nfs4_cb_procedures[op];
1199         cb->cb_msg.rpc_argp = cb;
1200         cb->cb_msg.rpc_resp = cb;
1201         cb->cb_ops = ops;
1202         INIT_WORK(&cb->cb_work, nfsd4_run_cb_work);
1203         cb->cb_seq_status = 1;
1204         cb->cb_status = 0;
1205         cb->cb_need_restart = false;
1206 }
1207
1208 void nfsd4_run_cb(struct nfsd4_callback *cb)
1209 {
1210         queue_work(callback_wq, &cb->cb_work);
1211 }