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/nfs4.h>
  35 #include <linux/sunrpc/clnt.h>
  36 #include <linux/sunrpc/xprt.h>
  37 #include <linux/sunrpc/svc_xprt.h>
  38 #include <linux/slab.h>
  39 #include "nfsd.h"
  40 #include "state.h"
  41 #include "netns.h"
  42 #include "trace.h"
  43 #include "xdr4cb.h"
  44 #include "xdr4.h"
  45
  46 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  47
  48 static void nfsd4_mark_cb_fault(struct nfs4_client *, int reason);
  49
  50 #define NFSPROC4_CB_NULL 0
  51 #define NFSPROC4_CB_COMPOUND 1
  52
  53 /* Index of predefined Linux callback client operations */
  54
  55 struct nfs4_cb_compound_hdr {
  56         /* args */
  57         u32             ident;  /* minorversion 0 only */
  58         u32             nops;
  59         __be32          *nops_p;
  60         u32             minorversion;
  61         /* res */
  62         int             status;
  63 };
  64
  65 static __be32 *xdr_encode_empty_array(__be32 *p)
  66 {
  67         *p++ = xdr_zero;
  68         return p;
  69 }
  70
  71 /*
  72  * Encode/decode NFSv4 CB basic data types
  73  *
  74  * Basic NFSv4 callback data types are defined in section 15 of RFC
  75  * 3530: "Network File System (NFS) version 4 Protocol" and section
  76  * 20 of RFC 5661: "Network File System (NFS) Version 4 Minor Version
  77  * 1 Protocol"
  78  */
  79
  80 static void encode_uint32(struct xdr_stream *xdr, u32 n)
  81 {
  82         WARN_ON_ONCE(xdr_stream_encode_u32(xdr, n) < 0);
  83 }
  84
  85 static void encode_bitmap4(struct xdr_stream *xdr, const __u32 *bitmap,
  86                            size_t len)
  87 {
  88         WARN_ON_ONCE(xdr_stream_encode_uint32_array(xdr, bitmap, len) < 0);
  89 }
  90
  91 static void encode_nfs_cb_opnum4(struct xdr_stream *xdr, enum nfs_cb_opnum4 op)
  92 {
  93         __be32 *p;
  94
  95         p = xdr_reserve_space(xdr, 4);
  96         *p = cpu_to_be32(op);
  97 }
  98
  99 /*
 100  * nfs_fh4
 101  *
 102  *      typedef opaque nfs_fh4<NFS4_FHSIZE>;
 103  */
 104 static void encode_nfs_fh4(struct xdr_stream *xdr, const struct knfsd_fh *fh)
 105 {
 106         u32 length = fh->fh_size;
 107         __be32 *p;
 108
 109         BUG_ON(length > NFS4_FHSIZE);
 110         p = xdr_reserve_space(xdr, 4 + length);
 111         xdr_encode_opaque(p, &fh->fh_raw, length);
 112 }
 113
 114 /*
 115  * stateid4
 116  *
 117  *      struct stateid4 {
 118  *              uint32_t        seqid;
 119  *              opaque          other[12];
 120  *      };
 121  */
 122 static void encode_stateid4(struct xdr_stream *xdr, const stateid_t *sid)
 123 {
 124         __be32 *p;
 125
 126         p = xdr_reserve_space(xdr, NFS4_STATEID_SIZE);
 127         *p++ = cpu_to_be32(sid->si_generation);
 128         xdr_encode_opaque_fixed(p, &sid->si_opaque, NFS4_STATEID_OTHER_SIZE);
 129 }
 130
 131 /*
 132  * sessionid4
 133  *
 134  *      typedef opaque sessionid4[NFS4_SESSIONID_SIZE];
 135  */
 136 static void encode_sessionid4(struct xdr_stream *xdr,
 137                               const struct nfsd4_session *session)
 138 {
 139         __be32 *p;
 140
 141         p = xdr_reserve_space(xdr, NFS4_MAX_SESSIONID_LEN);
 142         xdr_encode_opaque_fixed(p, session->se_sessionid.data,
 143                                         NFS4_MAX_SESSIONID_LEN);
 144 }
 145
 146 /*
 147  * nfsstat4
 148  */
 149 static const struct {
 150         int stat;
 151         int errno;
 152 } nfs_cb_errtbl[] = {
 153         { NFS4_OK,              0               },
 154         { NFS4ERR_PERM,         -EPERM          },
 155         { NFS4ERR_NOENT,        -ENOENT         },
 156         { NFS4ERR_IO,           -EIO            },
 157         { NFS4ERR_NXIO,         -ENXIO          },
 158         { NFS4ERR_ACCESS,       -EACCES         },
 159         { NFS4ERR_EXIST,        -EEXIST         },
 160         { NFS4ERR_XDEV,         -EXDEV          },
 161         { NFS4ERR_NOTDIR,       -ENOTDIR        },
 162         { NFS4ERR_ISDIR,        -EISDIR         },
 163         { NFS4ERR_INVAL,        -EINVAL         },
 164         { NFS4ERR_FBIG,         -EFBIG          },
 165         { NFS4ERR_NOSPC,        -ENOSPC         },
 166         { NFS4ERR_ROFS,         -EROFS          },
 167         { NFS4ERR_MLINK,        -EMLINK         },
 168         { NFS4ERR_NAMETOOLONG,  -ENAMETOOLONG   },
 169         { NFS4ERR_NOTEMPTY,     -ENOTEMPTY      },
 170         { NFS4ERR_DQUOT,        -EDQUOT         },
 171         { NFS4ERR_STALE,        -ESTALE         },
 172         { NFS4ERR_BADHANDLE,    -EBADHANDLE     },
 173         { NFS4ERR_BAD_COOKIE,   -EBADCOOKIE     },
 174         { NFS4ERR_NOTSUPP,      -ENOTSUPP       },
 175         { NFS4ERR_TOOSMALL,     -ETOOSMALL      },
 176         { NFS4ERR_SERVERFAULT,  -ESERVERFAULT   },
 177         { NFS4ERR_BADTYPE,      -EBADTYPE       },
 178         { NFS4ERR_LOCKED,       -EAGAIN         },
 179         { NFS4ERR_RESOURCE,     -EREMOTEIO      },
 180         { NFS4ERR_SYMLINK,      -ELOOP          },
 181         { NFS4ERR_OP_ILLEGAL,   -EOPNOTSUPP     },
 182         { NFS4ERR_DEADLOCK,     -EDEADLK        },
 183         { -1,                   -EIO            }
 184 };
 185
 186 /*
 187  * If we cannot translate the error, the recovery routines should
 188  * handle it.
 189  *
 190  * Note: remaining NFSv4 error codes have values > 10000, so should
 191  * not conflict with native Linux error codes.
 192  */
 193 static int nfs_cb_stat_to_errno(int status)
 194 {
 195         int i;
 196
 197         for (i = 0; nfs_cb_errtbl[i].stat != -1; i++) {
 198                 if (nfs_cb_errtbl[i].stat == status)
 199                         return nfs_cb_errtbl[i].errno;
 200         }
 201
 202         dprintk("NFSD: Unrecognized NFS CB status value: %u\n", status);
 203         return -status;
 204 }
 205
 206 static int decode_cb_op_status(struct xdr_stream *xdr,
 207                                enum nfs_cb_opnum4 expected, int *status)
 208 {
 209         __be32 *p;
 210         u32 op;
 211
 212         p = xdr_inline_decode(xdr, 4 + 4);
 213         if (unlikely(p == NULL))
 214                 goto out_overflow;
 215         op = be32_to_cpup(p++);
 216         if (unlikely(op != expected))
 217                 goto out_unexpected;
 218         *status = nfs_cb_stat_to_errno(be32_to_cpup(p));
 219         return 0;
 220 out_overflow:
 221         return -EIO;
 222 out_unexpected:
 223         dprintk("NFSD: Callback server returned operation %d but "
 224                 "we issued a request for %d\n", op, expected);
 225         return -EIO;
 226 }
 227
 228 /*
 229  * CB_COMPOUND4args
 230  *
 231  *      struct CB_COMPOUND4args {
 232  *              utf8str_cs      tag;
 233  *              uint32_t        minorversion;
 234  *              uint32_t        callback_ident;
 235  *              nfs_cb_argop4   argarray<>;
 236  *      };
 237 */
 238 static void encode_cb_compound4args(struct xdr_stream *xdr,
 239                                     struct nfs4_cb_compound_hdr *hdr)
 240 {
 241         __be32 * p;
 242
 243         p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4);
 244         p = xdr_encode_empty_array(p);          /* empty tag */
 245         *p++ = cpu_to_be32(hdr->minorversion);
 246         *p++ = cpu_to_be32(hdr->ident);
 247
 248         hdr->nops_p = p;
 249         *p = cpu_to_be32(hdr->nops);            /* argarray element count */
 250 }
 251
 252 /*
 253  * Update argarray element count
 254  */
 255 static void encode_cb_nops(struct nfs4_cb_compound_hdr *hdr)
 256 {
 257         BUG_ON(hdr->nops > NFS4_MAX_BACK_CHANNEL_OPS);
 258         *hdr->nops_p = cpu_to_be32(hdr->nops);
 259 }
 260
 261 /*
 262  * CB_COMPOUND4res
 263  *
 264  *      struct CB_COMPOUND4res {
 265  *              nfsstat4        status;
 266  *              utf8str_cs      tag;
 267  *              nfs_cb_resop4   resarray<>;
 268  *      };
 269  */
 270 static int decode_cb_compound4res(struct xdr_stream *xdr,
 271                                   struct nfs4_cb_compound_hdr *hdr)
 272 {
 273         u32 length;
 274         __be32 *p;
 275
 276         p = xdr_inline_decode(xdr, 4 + 4);
 277         if (unlikely(p == NULL))
 278                 goto out_overflow;
 279         hdr->status = be32_to_cpup(p++);
 280         /* Ignore the tag */
 281         length = be32_to_cpup(p++);
 282         p = xdr_inline_decode(xdr, length + 4);
 283         if (unlikely(p == NULL))
 284                 goto out_overflow;
 285         p += XDR_QUADLEN(length);
 286         hdr->nops = be32_to_cpup(p);
 287         return 0;
 288 out_overflow:
 289         return -EIO;
 290 }
 291
 292 /*
 293  * CB_RECALL4args
 294  *
 295  *      struct CB_RECALL4args {
 296  *              stateid4        stateid;
 297  *              bool            truncate;
 298  *              nfs_fh4         fh;
 299  *      };
 300  */
 301 static void encode_cb_recall4args(struct xdr_stream *xdr,
 302                                   const struct nfs4_delegation *dp,
 303                                   struct nfs4_cb_compound_hdr *hdr)
 304 {
 305         __be32 *p;
 306
 307         encode_nfs_cb_opnum4(xdr, OP_CB_RECALL);
 308         encode_stateid4(xdr, &dp->dl_stid.sc_stateid);
 309
 310         p = xdr_reserve_space(xdr, 4);
 311         *p++ = xdr_zero;                        /* truncate */
 312
 313         encode_nfs_fh4(xdr, &dp->dl_stid.sc_file->fi_fhandle);
 314
 315         hdr->nops++;
 316 }
 317
 318 /*
 319  * CB_RECALLANY4args
 320  *
 321  *      struct CB_RECALLANY4args {
 322  *              uint32_t        craa_objects_to_keep;
 323  *              bitmap4         craa_type_mask;
 324  *      };
 325  */
 326 static void
 327 encode_cb_recallany4args(struct xdr_stream *xdr,
 328         struct nfs4_cb_compound_hdr *hdr, struct nfsd4_cb_recall_any *ra)
 329 {
 330         encode_nfs_cb_opnum4(xdr, OP_CB_RECALL_ANY);
 331         encode_uint32(xdr, ra->ra_keep);
 332         encode_bitmap4(xdr, ra->ra_bmval, ARRAY_SIZE(ra->ra_bmval));
 333         hdr->nops++;
 334 }
 335
 336 /*
 337  * CB_SEQUENCE4args
 338  *
 339  *      struct CB_SEQUENCE4args {
 340  *              sessionid4              csa_sessionid;
 341  *              sequenceid4             csa_sequenceid;
 342  *              slotid4                 csa_slotid;
 343  *              slotid4                 csa_highest_slotid;
 344  *              bool                    csa_cachethis;
 345  *              referring_call_list4    csa_referring_call_lists<>;
 346  *      };
 347  */
 348 static void encode_cb_sequence4args(struct xdr_stream *xdr,
 349                                     const struct nfsd4_callback *cb,
 350                                     struct nfs4_cb_compound_hdr *hdr)
 351 {
 352         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
 353         __be32 *p;
 354
 355         if (hdr->minorversion == 0)
 356                 return;
 357
 358         encode_nfs_cb_opnum4(xdr, OP_CB_SEQUENCE);
 359         encode_sessionid4(xdr, session);
 360
 361         p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4 + 4);
 362         *p++ = cpu_to_be32(session->se_cb_seq_nr);      /* csa_sequenceid */
 363         *p++ = xdr_zero;                        /* csa_slotid */
 364         *p++ = xdr_zero;                        /* csa_highest_slotid */
 365         *p++ = xdr_zero;                        /* csa_cachethis */
 366         xdr_encode_empty_array(p);              /* csa_referring_call_lists */
 367
 368         hdr->nops++;
 369 }
 370
 371 /*
 372  * CB_SEQUENCE4resok
 373  *
 374  *      struct CB_SEQUENCE4resok {
 375  *              sessionid4      csr_sessionid;
 376  *              sequenceid4     csr_sequenceid;
 377  *              slotid4         csr_slotid;
 378  *              slotid4         csr_highest_slotid;
 379  *              slotid4         csr_target_highest_slotid;
 380  *      };
 381  *
 382  *      union CB_SEQUENCE4res switch (nfsstat4 csr_status) {
 383  *      case NFS4_OK:
 384  *              CB_SEQUENCE4resok       csr_resok4;
 385  *      default:
 386  *              void;
 387  *      };
 388  *
 389  * Our current back channel implmentation supports a single backchannel
 390  * with a single slot.
 391  */
 392 static int decode_cb_sequence4resok(struct xdr_stream *xdr,
 393                                     struct nfsd4_callback *cb)
 394 {
 395         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
 396         int status = -ESERVERFAULT;
 397         __be32 *p;
 398         u32 dummy;
 399
 400         /*
 401          * If the server returns different values for sessionID, slotID or
 402          * sequence number, the server is looney tunes.
 403          */
 404         p = xdr_inline_decode(xdr, NFS4_MAX_SESSIONID_LEN + 4 + 4 + 4 + 4);
 405         if (unlikely(p == NULL))
 406                 goto out_overflow;
 407
 408         if (memcmp(p, session->se_sessionid.data, NFS4_MAX_SESSIONID_LEN)) {
 409                 dprintk("NFS: %s Invalid session id\n", __func__);
 410                 goto out;
 411         }
 412         p += XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN);
 413
 414         dummy = be32_to_cpup(p++);
 415         if (dummy != session->se_cb_seq_nr) {
 416                 dprintk("NFS: %s Invalid sequence number\n", __func__);
 417                 goto out;
 418         }
 419
 420         dummy = be32_to_cpup(p++);
 421         if (dummy != 0) {
 422                 dprintk("NFS: %s Invalid slotid\n", __func__);
 423                 goto out;
 424         }
 425
 426         /*
 427          * FIXME: process highest slotid and target highest slotid
 428          */
 429         status = 0;
 430 out:
 431         cb->cb_seq_status = status;
 432         return status;
 433 out_overflow:
 434         status = -EIO;
 435         goto out;
 436 }
 437
 438 static int decode_cb_sequence4res(struct xdr_stream *xdr,
 439                                   struct nfsd4_callback *cb)
 440 {
 441         int status;
 442
 443         if (cb->cb_clp->cl_minorversion == 0)
 444                 return 0;
 445
 446         status = decode_cb_op_status(xdr, OP_CB_SEQUENCE, &cb->cb_seq_status);
 447         if (unlikely(status || cb->cb_seq_status))
 448                 return status;
 449
 450         return decode_cb_sequence4resok(xdr, cb);
 451 }
 452
 453 /*
 454  * NFSv4.0 and NFSv4.1 XDR encode functions
 455  *
 456  * NFSv4.0 callback argument types are defined in section 15 of RFC
 457  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
 458  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
 459  * Protocol".
 460  */
 461
 462 /*
 463  * NB: Without this zero space reservation, callbacks over krb5p fail
 464  */
 465 static void nfs4_xdr_enc_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
 466                                  const void *__unused)
 467 {
 468         xdr_reserve_space(xdr, 0);
 469 }
 470
 471 /*
 472  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
 473  */
 474 static void nfs4_xdr_enc_cb_recall(struct rpc_rqst *req, struct xdr_stream *xdr,
 475                                    const void *data)
 476 {
 477         const struct nfsd4_callback *cb = data;
 478         const struct nfs4_delegation *dp = cb_to_delegation(cb);
 479         struct nfs4_cb_compound_hdr hdr = {
 480                 .ident = cb->cb_clp->cl_cb_ident,
 481                 .minorversion = cb->cb_clp->cl_minorversion,
 482         };
 483
 484         encode_cb_compound4args(xdr, &hdr);
 485         encode_cb_sequence4args(xdr, cb, &hdr);
 486         encode_cb_recall4args(xdr, dp, &hdr);
 487         encode_cb_nops(&hdr);
 488 }
 489
 490 /*
 491  * 20.6. Operation 8: CB_RECALL_ANY - Keep Any N Recallable Objects
 492  */
 493 static void
 494 nfs4_xdr_enc_cb_recall_any(struct rpc_rqst *req,
 495                 struct xdr_stream *xdr, const void *data)
 496 {
 497         const struct nfsd4_callback *cb = data;
 498         struct nfsd4_cb_recall_any *ra;
 499         struct nfs4_cb_compound_hdr hdr = {
 500                 .ident = cb->cb_clp->cl_cb_ident,
 501                 .minorversion = cb->cb_clp->cl_minorversion,
 502         };
 503
 504         ra = container_of(cb, struct nfsd4_cb_recall_any, ra_cb);
 505         encode_cb_compound4args(xdr, &hdr);
 506         encode_cb_sequence4args(xdr, cb, &hdr);
 507         encode_cb_recallany4args(xdr, &hdr, ra);
 508         encode_cb_nops(&hdr);
 509 }
 510
 511 /*
 512  * NFSv4.0 and NFSv4.1 XDR decode functions
 513  *
 514  * NFSv4.0 callback result types are defined in section 15 of RFC
 515  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
 516  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
 517  * Protocol".
 518  */
 519
 520 static int nfs4_xdr_dec_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
 521                                 void *__unused)
 522 {
 523         return 0;
 524 }
 525
 526 /*
 527  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
 528  */
 529 static int nfs4_xdr_dec_cb_recall(struct rpc_rqst *rqstp,
 530                                   struct xdr_stream *xdr,
 531                                   void *data)
 532 {
 533         struct nfsd4_callback *cb = data;
 534         struct nfs4_cb_compound_hdr hdr;
 535         int status;
 536
 537         status = decode_cb_compound4res(xdr, &hdr);
 538         if (unlikely(status))
 539                 return status;
 540
 541         status = decode_cb_sequence4res(xdr, cb);
 542         if (unlikely(status || cb->cb_seq_status))
 543                 return status;
 544
 545         return decode_cb_op_status(xdr, OP_CB_RECALL, &cb->cb_status);
 546 }
 547
 548 /*
 549  * 20.6. Operation 8: CB_RECALL_ANY - Keep Any N Recallable Objects
 550  */
 551 static int
 552 nfs4_xdr_dec_cb_recall_any(struct rpc_rqst *rqstp,
 553                                   struct xdr_stream *xdr,
 554                                   void *data)
 555 {
 556         struct nfsd4_callback *cb = data;
 557         struct nfs4_cb_compound_hdr hdr;
 558         int status;
 559
 560         status = decode_cb_compound4res(xdr, &hdr);
 561         if (unlikely(status))
 562                 return status;
 563         status = decode_cb_sequence4res(xdr, cb);
 564         if (unlikely(status || cb->cb_seq_status))
 565                 return status;
 566         status =  decode_cb_op_status(xdr, OP_CB_RECALL_ANY, &cb->cb_status);
 567         return status;
 568 }
 569
 570 #ifdef CONFIG_NFSD_PNFS
 571 /*
 572  * CB_LAYOUTRECALL4args
 573  *
 574  *      struct layoutrecall_file4 {
 575  *              nfs_fh4         lor_fh;
 576  *              offset4         lor_offset;
 577  *              length4         lor_length;
 578  *              stateid4        lor_stateid;
 579  *      };
 580  *
 581  *      union layoutrecall4 switch(layoutrecall_type4 lor_recalltype) {
 582  *      case LAYOUTRECALL4_FILE:
 583  *              layoutrecall_file4 lor_layout;
 584  *      case LAYOUTRECALL4_FSID:
 585  *              fsid4              lor_fsid;
 586  *      case LAYOUTRECALL4_ALL:
 587  *              void;
 588  *      };
 589  *
 590  *      struct CB_LAYOUTRECALL4args {
 591  *              layouttype4             clora_type;
 592  *              layoutiomode4           clora_iomode;
 593  *              bool                    clora_changed;
 594  *              layoutrecall4           clora_recall;
 595  *      };
 596  */
 597 static void encode_cb_layout4args(struct xdr_stream *xdr,
 598                                   const struct nfs4_layout_stateid *ls,
 599                                   struct nfs4_cb_compound_hdr *hdr)
 600 {
 601         __be32 *p;
 602
 603         BUG_ON(hdr->minorversion == 0);
 604
 605         p = xdr_reserve_space(xdr, 5 * 4);
 606         *p++ = cpu_to_be32(OP_CB_LAYOUTRECALL);
 607         *p++ = cpu_to_be32(ls->ls_layout_type);
 608         *p++ = cpu_to_be32(IOMODE_ANY);
 609         *p++ = cpu_to_be32(1);
 610         *p = cpu_to_be32(RETURN_FILE);
 611
 612         encode_nfs_fh4(xdr, &ls->ls_stid.sc_file->fi_fhandle);
 613
 614         p = xdr_reserve_space(xdr, 2 * 8);
 615         p = xdr_encode_hyper(p, 0);
 616         xdr_encode_hyper(p, NFS4_MAX_UINT64);
 617
 618         encode_stateid4(xdr, &ls->ls_recall_sid);
 619
 620         hdr->nops++;
 621 }
 622
 623 static void nfs4_xdr_enc_cb_layout(struct rpc_rqst *req,
 624                                    struct xdr_stream *xdr,
 625                                    const void *data)
 626 {
 627         const struct nfsd4_callback *cb = data;
 628         const struct nfs4_layout_stateid *ls =
 629                 container_of(cb, struct nfs4_layout_stateid, ls_recall);
 630         struct nfs4_cb_compound_hdr hdr = {
 631                 .ident = 0,
 632                 .minorversion = cb->cb_clp->cl_minorversion,
 633         };
 634
 635         encode_cb_compound4args(xdr, &hdr);
 636         encode_cb_sequence4args(xdr, cb, &hdr);
 637         encode_cb_layout4args(xdr, ls, &hdr);
 638         encode_cb_nops(&hdr);
 639 }
 640
 641 static int nfs4_xdr_dec_cb_layout(struct rpc_rqst *rqstp,
 642                                   struct xdr_stream *xdr,
 643                                   void *data)
 644 {
 645         struct nfsd4_callback *cb = data;
 646         struct nfs4_cb_compound_hdr hdr;
 647         int status;
 648
 649         status = decode_cb_compound4res(xdr, &hdr);
 650         if (unlikely(status))
 651                 return status;
 652
 653         status = decode_cb_sequence4res(xdr, cb);
 654         if (unlikely(status || cb->cb_seq_status))
 655                 return status;
 656
 657         return decode_cb_op_status(xdr, OP_CB_LAYOUTRECALL, &cb->cb_status);
 658 }
 659 #endif /* CONFIG_NFSD_PNFS */
 660
 661 static void encode_stateowner(struct xdr_stream *xdr, struct nfs4_stateowner *so)
 662 {
 663         __be32  *p;
 664
 665         p = xdr_reserve_space(xdr, 8 + 4 + so->so_owner.len);
 666         p = xdr_encode_opaque_fixed(p, &so->so_client->cl_clientid, 8);
 667         xdr_encode_opaque(p, so->so_owner.data, so->so_owner.len);
 668 }
 669
 670 static void nfs4_xdr_enc_cb_notify_lock(struct rpc_rqst *req,
 671                                         struct xdr_stream *xdr,
 672                                         const void *data)
 673 {
 674         const struct nfsd4_callback *cb = data;
 675         const struct nfsd4_blocked_lock *nbl =
 676                 container_of(cb, struct nfsd4_blocked_lock, nbl_cb);
 677         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)nbl->nbl_lock.fl_owner;
 678         struct nfs4_cb_compound_hdr hdr = {
 679                 .ident = 0,
 680                 .minorversion = cb->cb_clp->cl_minorversion,
 681         };
 682
 683         __be32 *p;
 684
 685         BUG_ON(hdr.minorversion == 0);
 686
 687         encode_cb_compound4args(xdr, &hdr);
 688         encode_cb_sequence4args(xdr, cb, &hdr);
 689
 690         p = xdr_reserve_space(xdr, 4);
 691         *p = cpu_to_be32(OP_CB_NOTIFY_LOCK);
 692         encode_nfs_fh4(xdr, &nbl->nbl_fh);
 693         encode_stateowner(xdr, &lo->lo_owner);
 694         hdr.nops++;
 695
 696         encode_cb_nops(&hdr);
 697 }
 698
 699 static int nfs4_xdr_dec_cb_notify_lock(struct rpc_rqst *rqstp,
 700                                         struct xdr_stream *xdr,
 701                                         void *data)
 702 {
 703         struct nfsd4_callback *cb = data;
 704         struct nfs4_cb_compound_hdr hdr;
 705         int status;
 706
 707         status = decode_cb_compound4res(xdr, &hdr);
 708         if (unlikely(status))
 709                 return status;
 710
 711         status = decode_cb_sequence4res(xdr, cb);
 712         if (unlikely(status || cb->cb_seq_status))
 713                 return status;
 714
 715         return decode_cb_op_status(xdr, OP_CB_NOTIFY_LOCK, &cb->cb_status);
 716 }
 717
 718 /*
 719  * struct write_response4 {
 720  *      stateid4        wr_callback_id<1>;
 721  *      length4         wr_count;
 722  *      stable_how4     wr_committed;
 723  *      verifier4       wr_writeverf;
 724  * };
 725  * union offload_info4 switch (nfsstat4 coa_status) {
 726  *      case NFS4_OK:
 727  *              write_response4 coa_resok4;
 728  *      default:
 729  *              length4         coa_bytes_copied;
 730  * };
 731  * struct CB_OFFLOAD4args {
 732  *      nfs_fh4         coa_fh;
 733  *      stateid4        coa_stateid;
 734  *      offload_info4   coa_offload_info;
 735  * };
 736  */
 737 static void encode_offload_info4(struct xdr_stream *xdr,
 738                                  const struct nfsd4_cb_offload *cbo)
 739 {
 740         __be32 *p;
 741
 742         p = xdr_reserve_space(xdr, 4);
 743         *p = cbo->co_nfserr;
 744         switch (cbo->co_nfserr) {
 745         case nfs_ok:
 746                 p = xdr_reserve_space(xdr, 4 + 8 + 4 + NFS4_VERIFIER_SIZE);
 747                 p = xdr_encode_empty_array(p);
 748                 p = xdr_encode_hyper(p, cbo->co_res.wr_bytes_written);
 749                 *p++ = cpu_to_be32(cbo->co_res.wr_stable_how);
 750                 p = xdr_encode_opaque_fixed(p, cbo->co_res.wr_verifier.data,
 751                                             NFS4_VERIFIER_SIZE);
 752                 break;
 753         default:
 754                 p = xdr_reserve_space(xdr, 8);
 755                 /* We always return success if bytes were written */
 756                 p = xdr_encode_hyper(p, 0);
 757         }
 758 }
 759
 760 static void encode_cb_offload4args(struct xdr_stream *xdr,
 761                                    const struct nfsd4_cb_offload *cbo,
 762                                    struct nfs4_cb_compound_hdr *hdr)
 763 {
 764         __be32 *p;
 765
 766         p = xdr_reserve_space(xdr, 4);
 767         *p = cpu_to_be32(OP_CB_OFFLOAD);
 768         encode_nfs_fh4(xdr, &cbo->co_fh);
 769         encode_stateid4(xdr, &cbo->co_res.cb_stateid);
 770         encode_offload_info4(xdr, cbo);
 771
 772         hdr->nops++;
 773 }
 774
 775 static void nfs4_xdr_enc_cb_offload(struct rpc_rqst *req,
 776                                     struct xdr_stream *xdr,
 777                                     const void *data)
 778 {
 779         const struct nfsd4_callback *cb = data;
 780         const struct nfsd4_cb_offload *cbo =
 781                 container_of(cb, struct nfsd4_cb_offload, co_cb);
 782         struct nfs4_cb_compound_hdr hdr = {
 783                 .ident = 0,
 784                 .minorversion = cb->cb_clp->cl_minorversion,
 785         };
 786
 787         encode_cb_compound4args(xdr, &hdr);
 788         encode_cb_sequence4args(xdr, cb, &hdr);
 789         encode_cb_offload4args(xdr, cbo, &hdr);
 790         encode_cb_nops(&hdr);
 791 }
 792
 793 static int nfs4_xdr_dec_cb_offload(struct rpc_rqst *rqstp,
 794                                    struct xdr_stream *xdr,
 795                                    void *data)
 796 {
 797         struct nfsd4_callback *cb = data;
 798         struct nfs4_cb_compound_hdr hdr;
 799         int status;
 800
 801         status = decode_cb_compound4res(xdr, &hdr);
 802         if (unlikely(status))
 803                 return status;
 804
 805         status = decode_cb_sequence4res(xdr, cb);
 806         if (unlikely(status || cb->cb_seq_status))
 807                 return status;
 808
 809         return decode_cb_op_status(xdr, OP_CB_OFFLOAD, &cb->cb_status);
 810 }
 811 /*
 812  * RPC procedure tables
 813  */
 814 #define PROC(proc, call, argtype, restype)                              \
 815 [NFSPROC4_CLNT_##proc] = {                                              \
 816         .p_proc    = NFSPROC4_CB_##call,                                \
 817         .p_encode  = nfs4_xdr_enc_##argtype,            \
 818         .p_decode  = nfs4_xdr_dec_##restype,                            \
 819         .p_arglen  = NFS4_enc_##argtype##_sz,                           \
 820         .p_replen  = NFS4_dec_##restype##_sz,                           \
 821         .p_statidx = NFSPROC4_CB_##call,                                \
 822         .p_name    = #proc,                                             \
 823 }
 824
 825 static const struct rpc_procinfo nfs4_cb_procedures[] = {
 826         PROC(CB_NULL,   NULL,           cb_null,        cb_null),
 827         PROC(CB_RECALL, COMPOUND,       cb_recall,      cb_recall),
 828 #ifdef CONFIG_NFSD_PNFS
 829         PROC(CB_LAYOUT, COMPOUND,       cb_layout,      cb_layout),
 830 #endif
 831         PROC(CB_NOTIFY_LOCK,    COMPOUND,       cb_notify_lock, cb_notify_lock),
 832         PROC(CB_OFFLOAD,        COMPOUND,       cb_offload,     cb_offload),
 833         PROC(CB_RECALL_ANY,     COMPOUND,       cb_recall_any,  cb_recall_any),
 834 };
 835
 836 static unsigned int nfs4_cb_counts[ARRAY_SIZE(nfs4_cb_procedures)];
 837 static const struct rpc_version nfs_cb_version4 = {
 838 /*
 839  * Note on the callback rpc program version number: despite language in rfc
 840  * 5661 section 18.36.3 requiring servers to use 4 in this field, the
 841  * official xdr descriptions for both 4.0 and 4.1 specify version 1, and
 842  * in practice that appears to be what implementations use.  The section
 843  * 18.36.3 language is expected to be fixed in an erratum.
 844  */
 845         .number                 = 1,
 846         .nrprocs                = ARRAY_SIZE(nfs4_cb_procedures),
 847         .procs                  = nfs4_cb_procedures,
 848         .counts                 = nfs4_cb_counts,
 849 };
 850
 851 static const struct rpc_version *nfs_cb_version[2] = {
 852         [1] = &nfs_cb_version4,
 853 };
 854
 855 static const struct rpc_program cb_program;
 856
 857 static struct rpc_stat cb_stats = {
 858         .program                = &cb_program
 859 };
 860
 861 #define NFS4_CALLBACK 0x40000000
 862 static const struct rpc_program cb_program = {
 863         .name                   = "nfs4_cb",
 864         .number                 = NFS4_CALLBACK,
 865         .nrvers                 = ARRAY_SIZE(nfs_cb_version),
 866         .version                = nfs_cb_version,
 867         .stats                  = &cb_stats,
 868         .pipe_dir_name          = "nfsd4_cb",
 869 };
 870
 871 static int max_cb_time(struct net *net)
 872 {
 873         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 874
 875         /*
 876          * nfsd4_lease is set to at most one hour in __nfsd4_write_time,
 877          * so we can use 32-bit math on it. Warn if that assumption
 878          * ever stops being true.
 879          */
 880         if (WARN_ON_ONCE(nn->nfsd4_lease > 3600))
 881                 return 360 * HZ;
 882
 883         return max(((u32)nn->nfsd4_lease)/10, 1u) * HZ;
 884 }
 885
 886 static struct workqueue_struct *callback_wq;
 887
 888 static bool nfsd4_queue_cb(struct nfsd4_callback *cb)
 889 {
 890         return queue_delayed_work(callback_wq, &cb->cb_work, 0);
 891 }
 892
 893 static void nfsd4_queue_cb_delayed(struct nfsd4_callback *cb,
 894                                    unsigned long msecs)
 895 {
 896         queue_delayed_work(callback_wq, &cb->cb_work,
 897                            msecs_to_jiffies(msecs));
 898 }
 899
 900 static void nfsd41_cb_inflight_begin(struct nfs4_client *clp)
 901 {
 902         atomic_inc(&clp->cl_cb_inflight);
 903 }
 904
 905 static void nfsd41_cb_inflight_end(struct nfs4_client *clp)
 906 {
 907
 908         if (atomic_dec_and_test(&clp->cl_cb_inflight))
 909                 wake_up_var(&clp->cl_cb_inflight);
 910 }
 911
 912 static void nfsd41_cb_inflight_wait_complete(struct nfs4_client *clp)
 913 {
 914         wait_var_event(&clp->cl_cb_inflight,
 915                         !atomic_read(&clp->cl_cb_inflight));
 916 }
 917
 918 static const struct cred *get_backchannel_cred(struct nfs4_client *clp, struct rpc_clnt *client, struct nfsd4_session *ses)
 919 {
 920         if (clp->cl_minorversion == 0) {
 921                 client->cl_principal = clp->cl_cred.cr_targ_princ ?
 922                         clp->cl_cred.cr_targ_princ : "nfs";
 923
 924                 return get_cred(rpc_machine_cred());
 925         } else {
 926                 struct cred *kcred;
 927
 928                 kcred = prepare_kernel_cred(&init_task);
 929                 if (!kcred)
 930                         return NULL;
 931
 932                 kcred->fsuid = ses->se_cb_sec.uid;
 933                 kcred->fsgid = ses->se_cb_sec.gid;
 934                 return kcred;
 935         }
 936 }
 937
 938 static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn, struct nfsd4_session *ses)
 939 {
 940         int maxtime = max_cb_time(clp->net);
 941         struct rpc_timeout      timeparms = {
 942                 .to_initval     = maxtime,
 943                 .to_retries     = 0,
 944                 .to_maxval      = maxtime,
 945         };
 946         struct rpc_create_args args = {
 947                 .net            = clp->net,
 948                 .address        = (struct sockaddr *) &conn->cb_addr,
 949                 .addrsize       = conn->cb_addrlen,
 950                 .saddress       = (struct sockaddr *) &conn->cb_saddr,
 951                 .timeout        = &timeparms,
 952                 .program        = &cb_program,
 953                 .version        = 1,
 954                 .flags          = (RPC_CLNT_CREATE_NOPING | RPC_CLNT_CREATE_QUIET),
 955                 .cred           = current_cred(),
 956         };
 957         struct rpc_clnt *client;
 958         const struct cred *cred;
 959
 960         if (clp->cl_minorversion == 0) {
 961                 if (!clp->cl_cred.cr_principal &&
 962                     (clp->cl_cred.cr_flavor >= RPC_AUTH_GSS_KRB5)) {
 963                         trace_nfsd_cb_setup_err(clp, -EINVAL);
 964                         return -EINVAL;
 965                 }
 966                 args.client_name = clp->cl_cred.cr_principal;
 967                 args.prognumber = conn->cb_prog;
 968                 args.protocol = XPRT_TRANSPORT_TCP;
 969                 args.authflavor = clp->cl_cred.cr_flavor;
 970                 clp->cl_cb_ident = conn->cb_ident;
 971         } else {
 972                 if (!conn->cb_xprt)
 973                         return -EINVAL;
 974                 clp->cl_cb_session = ses;
 975                 args.bc_xprt = conn->cb_xprt;
 976                 args.prognumber = clp->cl_cb_session->se_cb_prog;
 977                 args.protocol = conn->cb_xprt->xpt_class->xcl_ident |
 978                                 XPRT_TRANSPORT_BC;
 979                 args.authflavor = ses->se_cb_sec.flavor;
 980         }
 981         /* Create RPC client */
 982         client = rpc_create(&args);
 983         if (IS_ERR(client)) {
 984                 trace_nfsd_cb_setup_err(clp, PTR_ERR(client));
 985                 return PTR_ERR(client);
 986         }
 987         cred = get_backchannel_cred(clp, client, ses);
 988         if (!cred) {
 989                 trace_nfsd_cb_setup_err(clp, -ENOMEM);
 990                 rpc_shutdown_client(client);
 991                 return -ENOMEM;
 992         }
 993
 994         if (clp->cl_minorversion != 0)
 995                 clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
 996         clp->cl_cb_client = client;
 997         clp->cl_cb_cred = cred;
 998         rcu_read_lock();
 999         trace_nfsd_cb_setup(clp, rpc_peeraddr2str(client, RPC_DISPLAY_NETID),
1000                             args.authflavor);
1001         rcu_read_unlock();
1002         return 0;
1003 }
1004
1005 static void nfsd4_mark_cb_state(struct nfs4_client *clp, int newstate)
1006 {
1007         if (clp->cl_cb_state != newstate) {
1008                 clp->cl_cb_state = newstate;
1009                 trace_nfsd_cb_state(clp);
1010         }
1011 }
1012
1013 static void nfsd4_mark_cb_down(struct nfs4_client *clp, int reason)
1014 {
1015         if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
1016                 return;
1017         nfsd4_mark_cb_state(clp, NFSD4_CB_DOWN);
1018 }
1019
1020 static void nfsd4_mark_cb_fault(struct nfs4_client *clp, int reason)
1021 {
1022         if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
1023                 return;
1024         nfsd4_mark_cb_state(clp, NFSD4_CB_FAULT);
1025 }
1026
1027 static void nfsd4_cb_probe_done(struct rpc_task *task, void *calldata)
1028 {
1029         struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
1030
1031         if (task->tk_status)
1032                 nfsd4_mark_cb_down(clp, task->tk_status);
1033         else
1034                 nfsd4_mark_cb_state(clp, NFSD4_CB_UP);
1035 }
1036
1037 static void nfsd4_cb_probe_release(void *calldata)
1038 {
1039         struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
1040
1041         nfsd41_cb_inflight_end(clp);
1042
1043 }
1044
1045 static const struct rpc_call_ops nfsd4_cb_probe_ops = {
1046         /* XXX: release method to ensure we set the cb channel down if
1047          * necessary on early failure? */
1048         .rpc_call_done = nfsd4_cb_probe_done,
1049         .rpc_release = nfsd4_cb_probe_release,
1050 };
1051
1052 /*
1053  * Poke the callback thread to process any updates to the callback
1054  * parameters, and send a null probe.
1055  */
1056 void nfsd4_probe_callback(struct nfs4_client *clp)
1057 {
1058         trace_nfsd_cb_probe(clp);
1059         nfsd4_mark_cb_state(clp, NFSD4_CB_UNKNOWN);
1060         set_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
1061         nfsd4_run_cb(&clp->cl_cb_null);
1062 }
1063
1064 void nfsd4_probe_callback_sync(struct nfs4_client *clp)
1065 {
1066         nfsd4_probe_callback(clp);
1067         flush_workqueue(callback_wq);
1068 }
1069
1070 void nfsd4_change_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
1071 {
1072         nfsd4_mark_cb_state(clp, NFSD4_CB_UNKNOWN);
1073         spin_lock(&clp->cl_lock);
1074         memcpy(&clp->cl_cb_conn, conn, sizeof(struct nfs4_cb_conn));
1075         spin_unlock(&clp->cl_lock);
1076 }
1077
1078 /*
1079  * There's currently a single callback channel slot.
1080  * If the slot is available, then mark it busy.  Otherwise, set the
1081  * thread for sleeping on the callback RPC wait queue.
1082  */
1083 static bool nfsd41_cb_get_slot(struct nfsd4_callback *cb, struct rpc_task *task)
1084 {
1085         struct nfs4_client *clp = cb->cb_clp;
1086
1087         if (!cb->cb_holds_slot &&
1088             test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
1089                 rpc_sleep_on(&clp->cl_cb_waitq, task, NULL);
1090                 /* Race breaker */
1091                 if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
1092                         dprintk("%s slot is busy\n", __func__);
1093                         return false;
1094                 }
1095                 rpc_wake_up_queued_task(&clp->cl_cb_waitq, task);
1096         }
1097         cb->cb_holds_slot = true;
1098         return true;
1099 }
1100
1101 static void nfsd41_cb_release_slot(struct nfsd4_callback *cb)
1102 {
1103         struct nfs4_client *clp = cb->cb_clp;
1104
1105         if (cb->cb_holds_slot) {
1106                 cb->cb_holds_slot = false;
1107                 clear_bit(0, &clp->cl_cb_slot_busy);
1108                 rpc_wake_up_next(&clp->cl_cb_waitq);
1109         }
1110 }
1111
1112 static void nfsd41_destroy_cb(struct nfsd4_callback *cb)
1113 {
1114         struct nfs4_client *clp = cb->cb_clp;
1115
1116         nfsd41_cb_release_slot(cb);
1117         if (cb->cb_ops && cb->cb_ops->release)
1118                 cb->cb_ops->release(cb);
1119         nfsd41_cb_inflight_end(clp);
1120 }
1121
1122 /*
1123  * TODO: cb_sequence should support referring call lists, cachethis, multiple
1124  * slots, and mark callback channel down on communication errors.
1125  */
1126 static void nfsd4_cb_prepare(struct rpc_task *task, void *calldata)
1127 {
1128         struct nfsd4_callback *cb = calldata;
1129         struct nfs4_client *clp = cb->cb_clp;
1130         u32 minorversion = clp->cl_minorversion;
1131
1132         /*
1133          * cb_seq_status is only set in decode_cb_sequence4res,
1134          * and so will remain 1 if an rpc level failure occurs.
1135          */
1136         cb->cb_seq_status = 1;
1137         cb->cb_status = 0;
1138         if (minorversion && !nfsd41_cb_get_slot(cb, task))
1139                 return;
1140         rpc_call_start(task);
1141 }
1142
1143 static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback *cb)
1144 {
1145         struct nfs4_client *clp = cb->cb_clp;
1146         struct nfsd4_session *session = clp->cl_cb_session;
1147         bool ret = true;
1148
1149         if (!clp->cl_minorversion) {
1150                 /*
1151                  * If the backchannel connection was shut down while this
1152                  * task was queued, we need to resubmit it after setting up
1153                  * a new backchannel connection.
1154                  *
1155                  * Note that if we lost our callback connection permanently
1156                  * the submission code will error out, so we don't need to
1157                  * handle that case here.
1158                  */
1159                 if (RPC_SIGNALLED(task))
1160                         goto need_restart;
1161
1162                 return true;
1163         }
1164
1165         if (!cb->cb_holds_slot)
1166                 goto need_restart;
1167
1168         switch (cb->cb_seq_status) {
1169         case 0:
1170                 /*
1171                  * No need for lock, access serialized in nfsd4_cb_prepare
1172                  *
1173                  * RFC5661 20.9.3
1174                  * If CB_SEQUENCE returns an error, then the state of the slot
1175                  * (sequence ID, cached reply) MUST NOT change.
1176                  */
1177                 ++session->se_cb_seq_nr;
1178                 break;
1179         case -ESERVERFAULT:
1180                 ++session->se_cb_seq_nr;
1181                 nfsd4_mark_cb_fault(cb->cb_clp, cb->cb_seq_status);
1182                 ret = false;
1183                 break;
1184         case 1:
1185                 /*
1186                  * cb_seq_status remains 1 if an RPC Reply was never
1187                  * received. NFSD can't know if the client processed
1188                  * the CB_SEQUENCE operation. Ask the client to send a
1189                  * DESTROY_SESSION to recover.
1190                  */
1191                 fallthrough;
1192         case -NFS4ERR_BADSESSION:
1193                 nfsd4_mark_cb_fault(cb->cb_clp, cb->cb_seq_status);
1194                 ret = false;
1195                 goto need_restart;
1196         case -NFS4ERR_DELAY:
1197                 cb->cb_seq_status = 1;
1198                 if (!rpc_restart_call(task))
1199                         goto out;
1200
1201                 rpc_delay(task, 2 * HZ);
1202                 return false;
1203         case -NFS4ERR_BADSLOT:
1204                 goto retry_nowait;
1205         case -NFS4ERR_SEQ_MISORDERED:
1206                 if (session->se_cb_seq_nr != 1) {
1207                         session->se_cb_seq_nr = 1;
1208                         goto retry_nowait;
1209                 }
1210                 break;
1211         default:
1212                 nfsd4_mark_cb_fault(cb->cb_clp, cb->cb_seq_status);
1213                 dprintk("%s: unprocessed error %d\n", __func__,
1214                         cb->cb_seq_status);
1215         }
1216
1217         nfsd41_cb_release_slot(cb);
1218         dprintk("%s: freed slot, new seqid=%d\n", __func__,
1219                 clp->cl_cb_session->se_cb_seq_nr);
1220
1221         if (RPC_SIGNALLED(task))
1222                 goto need_restart;
1223 out:
1224         return ret;
1225 retry_nowait:
1226         if (rpc_restart_call_prepare(task))
1227                 ret = false;
1228         goto out;
1229 need_restart:
1230         if (!test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags)) {
1231                 task->tk_status = 0;
1232                 cb->cb_need_restart = true;
1233         }
1234         return false;
1235 }
1236
1237 static void nfsd4_cb_done(struct rpc_task *task, void *calldata)
1238 {
1239         struct nfsd4_callback *cb = calldata;
1240         struct nfs4_client *clp = cb->cb_clp;
1241
1242         if (!nfsd4_cb_sequence_done(task, cb))
1243                 return;
1244
1245         if (cb->cb_status) {
1246                 WARN_ON_ONCE(task->tk_status);
1247                 task->tk_status = cb->cb_status;
1248         }
1249
1250         switch (cb->cb_ops->done(cb, task)) {
1251         case 0:
1252                 task->tk_status = 0;
1253                 rpc_restart_call_prepare(task);
1254                 return;
1255         case 1:
1256                 switch (task->tk_status) {
1257                 case -EIO:
1258                 case -ETIMEDOUT:
1259                 case -EACCES:
1260                         nfsd4_mark_cb_down(clp, task->tk_status);
1261                 }
1262                 break;
1263         default:
1264                 BUG();
1265         }
1266 }
1267
1268 static void nfsd4_cb_release(void *calldata)
1269 {
1270         struct nfsd4_callback *cb = calldata;
1271
1272         if (cb->cb_need_restart)
1273                 nfsd4_queue_cb(cb);
1274         else
1275                 nfsd41_destroy_cb(cb);
1276
1277 }
1278
1279 static const struct rpc_call_ops nfsd4_cb_ops = {
1280         .rpc_call_prepare = nfsd4_cb_prepare,
1281         .rpc_call_done = nfsd4_cb_done,
1282         .rpc_release = nfsd4_cb_release,
1283 };
1284
1285 int nfsd4_create_callback_queue(void)
1286 {
1287         callback_wq = alloc_ordered_workqueue("nfsd4_callbacks", 0);
1288         if (!callback_wq)
1289                 return -ENOMEM;
1290         return 0;
1291 }
1292
1293 void nfsd4_destroy_callback_queue(void)
1294 {
1295         destroy_workqueue(callback_wq);
1296 }
1297
1298 /* must be called under the state lock */
1299 void nfsd4_shutdown_callback(struct nfs4_client *clp)
1300 {
1301         if (clp->cl_cb_state != NFSD4_CB_UNKNOWN)
1302                 trace_nfsd_cb_shutdown(clp);
1303
1304         set_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags);
1305         /*
1306          * Note this won't actually result in a null callback;
1307          * instead, nfsd4_run_cb_null() will detect the killed
1308          * client, destroy the rpc client, and stop:
1309          */
1310         nfsd4_run_cb(&clp->cl_cb_null);
1311         flush_workqueue(callback_wq);
1312         nfsd41_cb_inflight_wait_complete(clp);
1313 }
1314
1315 /* requires cl_lock: */
1316 static struct nfsd4_conn * __nfsd4_find_backchannel(struct nfs4_client *clp)
1317 {
1318         struct nfsd4_session *s;
1319         struct nfsd4_conn *c;
1320
1321         list_for_each_entry(s, &clp->cl_sessions, se_perclnt) {
1322                 list_for_each_entry(c, &s->se_conns, cn_persession) {
1323                         if (c->cn_flags & NFS4_CDFC4_BACK)
1324                                 return c;
1325                 }
1326         }
1327         return NULL;
1328 }
1329
1330 /*
1331  * Note there isn't a lot of locking in this code; instead we depend on
1332  * the fact that it is run from the callback_wq, which won't run two
1333  * work items at once.  So, for example, callback_wq handles all access
1334  * of cl_cb_client and all calls to rpc_create or rpc_shutdown_client.
1335  */
1336 static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
1337 {
1338         struct nfs4_cb_conn conn;
1339         struct nfs4_client *clp = cb->cb_clp;
1340         struct nfsd4_session *ses = NULL;
1341         struct nfsd4_conn *c;
1342         int err;
1343
1344         /*
1345          * This is either an update, or the client dying; in either case,
1346          * kill the old client:
1347          */
1348         if (clp->cl_cb_client) {
1349                 rpc_shutdown_client(clp->cl_cb_client);
1350                 clp->cl_cb_client = NULL;
1351                 put_cred(clp->cl_cb_cred);
1352                 clp->cl_cb_cred = NULL;
1353         }
1354         if (clp->cl_cb_conn.cb_xprt) {
1355                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1356                 clp->cl_cb_conn.cb_xprt = NULL;
1357         }
1358         if (test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags))
1359                 return;
1360         spin_lock(&clp->cl_lock);
1361         /*
1362          * Only serialized callback code is allowed to clear these
1363          * flags; main nfsd code can only set them:
1364          */
1365         BUG_ON(!(clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK));
1366         clear_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
1367         memcpy(&conn, &cb->cb_clp->cl_cb_conn, sizeof(struct nfs4_cb_conn));
1368         c = __nfsd4_find_backchannel(clp);
1369         if (c) {
1370                 svc_xprt_get(c->cn_xprt);
1371                 conn.cb_xprt = c->cn_xprt;
1372                 ses = c->cn_session;
1373         }
1374         spin_unlock(&clp->cl_lock);
1375
1376         err = setup_callback_client(clp, &conn, ses);
1377         if (err) {
1378                 nfsd4_mark_cb_down(clp, err);
1379                 if (c)
1380                         svc_xprt_put(c->cn_xprt);
1381                 return;
1382         }
1383 }
1384
1385 static void
1386 nfsd4_run_cb_work(struct work_struct *work)
1387 {
1388         struct nfsd4_callback *cb =
1389                 container_of(work, struct nfsd4_callback, cb_work.work);
1390         struct nfs4_client *clp = cb->cb_clp;
1391         struct rpc_clnt *clnt;
1392         int flags;
1393
1394         if (clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK)
1395                 nfsd4_process_cb_update(cb);
1396
1397         clnt = clp->cl_cb_client;
1398         if (!clnt) {
1399                 if (test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags))
1400                         nfsd41_destroy_cb(cb);
1401                 else {
1402                         /*
1403                          * XXX: Ideally, we could wait for the client to
1404                          *      reconnect, but I haven't figured out how
1405                          *      to do that yet.
1406                          */
1407                         nfsd4_queue_cb_delayed(cb, 25);
1408                 }
1409                 return;
1410         }
1411
1412         /*
1413          * Don't send probe messages for 4.1 or later.
1414          */
1415         if (!cb->cb_ops && clp->cl_minorversion) {
1416                 nfsd4_mark_cb_state(clp, NFSD4_CB_UP);
1417                 nfsd41_destroy_cb(cb);
1418                 return;
1419         }
1420
1421         if (cb->cb_need_restart) {
1422                 cb->cb_need_restart = false;
1423         } else {
1424                 if (cb->cb_ops && cb->cb_ops->prepare)
1425                         cb->cb_ops->prepare(cb);
1426         }
1427         cb->cb_msg.rpc_cred = clp->cl_cb_cred;
1428         flags = clp->cl_minorversion ? RPC_TASK_NOCONNECT : RPC_TASK_SOFTCONN;
1429         rpc_call_async(clnt, &cb->cb_msg, RPC_TASK_SOFT | flags,
1430                         cb->cb_ops ? &nfsd4_cb_ops : &nfsd4_cb_probe_ops, cb);
1431 }
1432
1433 void nfsd4_init_cb(struct nfsd4_callback *cb, struct nfs4_client *clp,
1434                 const struct nfsd4_callback_ops *ops, enum nfsd4_cb_op op)
1435 {
1436         cb->cb_clp = clp;
1437         cb->cb_msg.rpc_proc = &nfs4_cb_procedures[op];
1438         cb->cb_msg.rpc_argp = cb;
1439         cb->cb_msg.rpc_resp = cb;
1440         cb->cb_ops = ops;
1441         INIT_DELAYED_WORK(&cb->cb_work, nfsd4_run_cb_work);
1442         cb->cb_seq_status = 1;
1443         cb->cb_status = 0;
1444         cb->cb_need_restart = false;
1445         cb->cb_holds_slot = false;
1446 }
1447
1448 /**
1449  * nfsd4_run_cb - queue up a callback job to run
1450  * @cb: callback to queue
1451  *
1452  * Kick off a callback to do its thing. Returns false if it was already
1453  * on a queue, true otherwise.
1454  */
1455 bool nfsd4_run_cb(struct nfsd4_callback *cb)
1456 {
1457         struct nfs4_client *clp = cb->cb_clp;
1458         bool queued;
1459
1460         nfsd41_cb_inflight_begin(clp);
1461         queued = nfsd4_queue_cb(cb);
1462         if (!queued)
1463                 nfsd41_cb_inflight_end(clp);
1464         return queued;
1465 }