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 *clp);
  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         xdr_stream_encode_uint32_array(xdr, bitmap, len);
  89 }
  90
  91 static int decode_cb_fattr4(struct xdr_stream *xdr, uint32_t *bitmap,
  92                                 struct nfs4_cb_fattr *fattr)
  93 {
  94         fattr->ncf_cb_change = 0;
  95         fattr->ncf_cb_fsize = 0;
  96         if (bitmap[0] & FATTR4_WORD0_CHANGE)
  97                 if (xdr_stream_decode_u64(xdr, &fattr->ncf_cb_change) < 0)
  98                         return -NFSERR_BAD_XDR;
  99         if (bitmap[0] & FATTR4_WORD0_SIZE)
 100                 if (xdr_stream_decode_u64(xdr, &fattr->ncf_cb_fsize) < 0)
 101                         return -NFSERR_BAD_XDR;
 102         return 0;
 103 }
 104
 105 static void encode_nfs_cb_opnum4(struct xdr_stream *xdr, enum nfs_cb_opnum4 op)
 106 {
 107         __be32 *p;
 108
 109         p = xdr_reserve_space(xdr, 4);
 110         *p = cpu_to_be32(op);
 111 }
 112
 113 /*
 114  * nfs_fh4
 115  *
 116  *      typedef opaque nfs_fh4<NFS4_FHSIZE>;
 117  */
 118 static void encode_nfs_fh4(struct xdr_stream *xdr, const struct knfsd_fh *fh)
 119 {
 120         u32 length = fh->fh_size;
 121         __be32 *p;
 122
 123         BUG_ON(length > NFS4_FHSIZE);
 124         p = xdr_reserve_space(xdr, 4 + length);
 125         xdr_encode_opaque(p, &fh->fh_raw, length);
 126 }
 127
 128 /*
 129  * stateid4
 130  *
 131  *      struct stateid4 {
 132  *              uint32_t        seqid;
 133  *              opaque          other[12];
 134  *      };
 135  */
 136 static void encode_stateid4(struct xdr_stream *xdr, const stateid_t *sid)
 137 {
 138         __be32 *p;
 139
 140         p = xdr_reserve_space(xdr, NFS4_STATEID_SIZE);
 141         *p++ = cpu_to_be32(sid->si_generation);
 142         xdr_encode_opaque_fixed(p, &sid->si_opaque, NFS4_STATEID_OTHER_SIZE);
 143 }
 144
 145 /*
 146  * sessionid4
 147  *
 148  *      typedef opaque sessionid4[NFS4_SESSIONID_SIZE];
 149  */
 150 static void encode_sessionid4(struct xdr_stream *xdr,
 151                               const struct nfsd4_session *session)
 152 {
 153         __be32 *p;
 154
 155         p = xdr_reserve_space(xdr, NFS4_MAX_SESSIONID_LEN);
 156         xdr_encode_opaque_fixed(p, session->se_sessionid.data,
 157                                         NFS4_MAX_SESSIONID_LEN);
 158 }
 159
 160 /*
 161  * nfsstat4
 162  */
 163 static const struct {
 164         int stat;
 165         int errno;
 166 } nfs_cb_errtbl[] = {
 167         { NFS4_OK,              0               },
 168         { NFS4ERR_PERM,         -EPERM          },
 169         { NFS4ERR_NOENT,        -ENOENT         },
 170         { NFS4ERR_IO,           -EIO            },
 171         { NFS4ERR_NXIO,         -ENXIO          },
 172         { NFS4ERR_ACCESS,       -EACCES         },
 173         { NFS4ERR_EXIST,        -EEXIST         },
 174         { NFS4ERR_XDEV,         -EXDEV          },
 175         { NFS4ERR_NOTDIR,       -ENOTDIR        },
 176         { NFS4ERR_ISDIR,        -EISDIR         },
 177         { NFS4ERR_INVAL,        -EINVAL         },
 178         { NFS4ERR_FBIG,         -EFBIG          },
 179         { NFS4ERR_NOSPC,        -ENOSPC         },
 180         { NFS4ERR_ROFS,         -EROFS          },
 181         { NFS4ERR_MLINK,        -EMLINK         },
 182         { NFS4ERR_NAMETOOLONG,  -ENAMETOOLONG   },
 183         { NFS4ERR_NOTEMPTY,     -ENOTEMPTY      },
 184         { NFS4ERR_DQUOT,        -EDQUOT         },
 185         { NFS4ERR_STALE,        -ESTALE         },
 186         { NFS4ERR_BADHANDLE,    -EBADHANDLE     },
 187         { NFS4ERR_BAD_COOKIE,   -EBADCOOKIE     },
 188         { NFS4ERR_NOTSUPP,      -ENOTSUPP       },
 189         { NFS4ERR_TOOSMALL,     -ETOOSMALL      },
 190         { NFS4ERR_SERVERFAULT,  -ESERVERFAULT   },
 191         { NFS4ERR_BADTYPE,      -EBADTYPE       },
 192         { NFS4ERR_LOCKED,       -EAGAIN         },
 193         { NFS4ERR_RESOURCE,     -EREMOTEIO      },
 194         { NFS4ERR_SYMLINK,      -ELOOP          },
 195         { NFS4ERR_OP_ILLEGAL,   -EOPNOTSUPP     },
 196         { NFS4ERR_DEADLOCK,     -EDEADLK        },
 197         { -1,                   -EIO            }
 198 };
 199
 200 /*
 201  * If we cannot translate the error, the recovery routines should
 202  * handle it.
 203  *
 204  * Note: remaining NFSv4 error codes have values > 10000, so should
 205  * not conflict with native Linux error codes.
 206  */
 207 static int nfs_cb_stat_to_errno(int status)
 208 {
 209         int i;
 210
 211         for (i = 0; nfs_cb_errtbl[i].stat != -1; i++) {
 212                 if (nfs_cb_errtbl[i].stat == status)
 213                         return nfs_cb_errtbl[i].errno;
 214         }
 215
 216         dprintk("NFSD: Unrecognized NFS CB status value: %u\n", status);
 217         return -status;
 218 }
 219
 220 static int decode_cb_op_status(struct xdr_stream *xdr,
 221                                enum nfs_cb_opnum4 expected, int *status)
 222 {
 223         __be32 *p;
 224         u32 op;
 225
 226         p = xdr_inline_decode(xdr, 4 + 4);
 227         if (unlikely(p == NULL))
 228                 goto out_overflow;
 229         op = be32_to_cpup(p++);
 230         if (unlikely(op != expected))
 231                 goto out_unexpected;
 232         *status = nfs_cb_stat_to_errno(be32_to_cpup(p));
 233         return 0;
 234 out_overflow:
 235         return -EIO;
 236 out_unexpected:
 237         dprintk("NFSD: Callback server returned operation %d but "
 238                 "we issued a request for %d\n", op, expected);
 239         return -EIO;
 240 }
 241
 242 /*
 243  * CB_COMPOUND4args
 244  *
 245  *      struct CB_COMPOUND4args {
 246  *              utf8str_cs      tag;
 247  *              uint32_t        minorversion;
 248  *              uint32_t        callback_ident;
 249  *              nfs_cb_argop4   argarray<>;
 250  *      };
 251 */
 252 static void encode_cb_compound4args(struct xdr_stream *xdr,
 253                                     struct nfs4_cb_compound_hdr *hdr)
 254 {
 255         __be32 * p;
 256
 257         p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4);
 258         p = xdr_encode_empty_array(p);          /* empty tag */
 259         *p++ = cpu_to_be32(hdr->minorversion);
 260         *p++ = cpu_to_be32(hdr->ident);
 261
 262         hdr->nops_p = p;
 263         *p = cpu_to_be32(hdr->nops);            /* argarray element count */
 264 }
 265
 266 /*
 267  * Update argarray element count
 268  */
 269 static void encode_cb_nops(struct nfs4_cb_compound_hdr *hdr)
 270 {
 271         BUG_ON(hdr->nops > NFS4_MAX_BACK_CHANNEL_OPS);
 272         *hdr->nops_p = cpu_to_be32(hdr->nops);
 273 }
 274
 275 /*
 276  * CB_COMPOUND4res
 277  *
 278  *      struct CB_COMPOUND4res {
 279  *              nfsstat4        status;
 280  *              utf8str_cs      tag;
 281  *              nfs_cb_resop4   resarray<>;
 282  *      };
 283  */
 284 static int decode_cb_compound4res(struct xdr_stream *xdr,
 285                                   struct nfs4_cb_compound_hdr *hdr)
 286 {
 287         u32 length;
 288         __be32 *p;
 289
 290         p = xdr_inline_decode(xdr, 4 + 4);
 291         if (unlikely(p == NULL))
 292                 goto out_overflow;
 293         hdr->status = be32_to_cpup(p++);
 294         /* Ignore the tag */
 295         length = be32_to_cpup(p++);
 296         p = xdr_inline_decode(xdr, length + 4);
 297         if (unlikely(p == NULL))
 298                 goto out_overflow;
 299         p += XDR_QUADLEN(length);
 300         hdr->nops = be32_to_cpup(p);
 301         return 0;
 302 out_overflow:
 303         return -EIO;
 304 }
 305
 306 /*
 307  * CB_RECALL4args
 308  *
 309  *      struct CB_RECALL4args {
 310  *              stateid4        stateid;
 311  *              bool            truncate;
 312  *              nfs_fh4         fh;
 313  *      };
 314  */
 315 static void encode_cb_recall4args(struct xdr_stream *xdr,
 316                                   const struct nfs4_delegation *dp,
 317                                   struct nfs4_cb_compound_hdr *hdr)
 318 {
 319         __be32 *p;
 320
 321         encode_nfs_cb_opnum4(xdr, OP_CB_RECALL);
 322         encode_stateid4(xdr, &dp->dl_stid.sc_stateid);
 323
 324         p = xdr_reserve_space(xdr, 4);
 325         *p++ = xdr_zero;                        /* truncate */
 326
 327         encode_nfs_fh4(xdr, &dp->dl_stid.sc_file->fi_fhandle);
 328
 329         hdr->nops++;
 330 }
 331
 332 /*
 333  * CB_RECALLANY4args
 334  *
 335  *      struct CB_RECALLANY4args {
 336  *              uint32_t        craa_objects_to_keep;
 337  *              bitmap4         craa_type_mask;
 338  *      };
 339  */
 340 static void
 341 encode_cb_recallany4args(struct xdr_stream *xdr,
 342         struct nfs4_cb_compound_hdr *hdr, struct nfsd4_cb_recall_any *ra)
 343 {
 344         encode_nfs_cb_opnum4(xdr, OP_CB_RECALL_ANY);
 345         encode_uint32(xdr, ra->ra_keep);
 346         encode_bitmap4(xdr, ra->ra_bmval, ARRAY_SIZE(ra->ra_bmval));
 347         hdr->nops++;
 348 }
 349
 350 /*
 351  * CB_GETATTR4args
 352  *      struct CB_GETATTR4args {
 353  *         nfs_fh4 fh;
 354  *         bitmap4 attr_request;
 355  *      };
 356  *
 357  * The size and change attributes are the only one
 358  * guaranteed to be serviced by the client.
 359  */
 360 static void
 361 encode_cb_getattr4args(struct xdr_stream *xdr, struct nfs4_cb_compound_hdr *hdr,
 362                         struct nfs4_cb_fattr *fattr)
 363 {
 364         struct nfs4_delegation *dp =
 365                 container_of(fattr, struct nfs4_delegation, dl_cb_fattr);
 366         struct knfsd_fh *fh = &dp->dl_stid.sc_file->fi_fhandle;
 367
 368         encode_nfs_cb_opnum4(xdr, OP_CB_GETATTR);
 369         encode_nfs_fh4(xdr, fh);
 370         encode_bitmap4(xdr, fattr->ncf_cb_bmap, ARRAY_SIZE(fattr->ncf_cb_bmap));
 371         hdr->nops++;
 372 }
 373
 374 /*
 375  * CB_SEQUENCE4args
 376  *
 377  *      struct CB_SEQUENCE4args {
 378  *              sessionid4              csa_sessionid;
 379  *              sequenceid4             csa_sequenceid;
 380  *              slotid4                 csa_slotid;
 381  *              slotid4                 csa_highest_slotid;
 382  *              bool                    csa_cachethis;
 383  *              referring_call_list4    csa_referring_call_lists<>;
 384  *      };
 385  */
 386 static void encode_cb_sequence4args(struct xdr_stream *xdr,
 387                                     const struct nfsd4_callback *cb,
 388                                     struct nfs4_cb_compound_hdr *hdr)
 389 {
 390         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
 391         __be32 *p;
 392
 393         if (hdr->minorversion == 0)
 394                 return;
 395
 396         encode_nfs_cb_opnum4(xdr, OP_CB_SEQUENCE);
 397         encode_sessionid4(xdr, session);
 398
 399         p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4 + 4);
 400         *p++ = cpu_to_be32(session->se_cb_seq_nr);      /* csa_sequenceid */
 401         *p++ = xdr_zero;                        /* csa_slotid */
 402         *p++ = xdr_zero;                        /* csa_highest_slotid */
 403         *p++ = xdr_zero;                        /* csa_cachethis */
 404         xdr_encode_empty_array(p);              /* csa_referring_call_lists */
 405
 406         hdr->nops++;
 407 }
 408
 409 /*
 410  * CB_SEQUENCE4resok
 411  *
 412  *      struct CB_SEQUENCE4resok {
 413  *              sessionid4      csr_sessionid;
 414  *              sequenceid4     csr_sequenceid;
 415  *              slotid4         csr_slotid;
 416  *              slotid4         csr_highest_slotid;
 417  *              slotid4         csr_target_highest_slotid;
 418  *      };
 419  *
 420  *      union CB_SEQUENCE4res switch (nfsstat4 csr_status) {
 421  *      case NFS4_OK:
 422  *              CB_SEQUENCE4resok       csr_resok4;
 423  *      default:
 424  *              void;
 425  *      };
 426  *
 427  * Our current back channel implmentation supports a single backchannel
 428  * with a single slot.
 429  */
 430 static int decode_cb_sequence4resok(struct xdr_stream *xdr,
 431                                     struct nfsd4_callback *cb)
 432 {
 433         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
 434         int status = -ESERVERFAULT;
 435         __be32 *p;
 436         u32 dummy;
 437
 438         /*
 439          * If the server returns different values for sessionID, slotID or
 440          * sequence number, the server is looney tunes.
 441          */
 442         p = xdr_inline_decode(xdr, NFS4_MAX_SESSIONID_LEN + 4 + 4 + 4 + 4);
 443         if (unlikely(p == NULL))
 444                 goto out_overflow;
 445
 446         if (memcmp(p, session->se_sessionid.data, NFS4_MAX_SESSIONID_LEN)) {
 447                 dprintk("NFS: %s Invalid session id\n", __func__);
 448                 goto out;
 449         }
 450         p += XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN);
 451
 452         dummy = be32_to_cpup(p++);
 453         if (dummy != session->se_cb_seq_nr) {
 454                 dprintk("NFS: %s Invalid sequence number\n", __func__);
 455                 goto out;
 456         }
 457
 458         dummy = be32_to_cpup(p++);
 459         if (dummy != 0) {
 460                 dprintk("NFS: %s Invalid slotid\n", __func__);
 461                 goto out;
 462         }
 463
 464         /*
 465          * FIXME: process highest slotid and target highest slotid
 466          */
 467         status = 0;
 468 out:
 469         cb->cb_seq_status = status;
 470         return status;
 471 out_overflow:
 472         status = -EIO;
 473         goto out;
 474 }
 475
 476 static int decode_cb_sequence4res(struct xdr_stream *xdr,
 477                                   struct nfsd4_callback *cb)
 478 {
 479         int status;
 480
 481         if (cb->cb_clp->cl_minorversion == 0)
 482                 return 0;
 483
 484         status = decode_cb_op_status(xdr, OP_CB_SEQUENCE, &cb->cb_seq_status);
 485         if (unlikely(status || cb->cb_seq_status))
 486                 return status;
 487
 488         return decode_cb_sequence4resok(xdr, cb);
 489 }
 490
 491 /*
 492  * NFSv4.0 and NFSv4.1 XDR encode functions
 493  *
 494  * NFSv4.0 callback argument types are defined in section 15 of RFC
 495  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
 496  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
 497  * Protocol".
 498  */
 499
 500 /*
 501  * NB: Without this zero space reservation, callbacks over krb5p fail
 502  */
 503 static void nfs4_xdr_enc_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
 504                                  const void *__unused)
 505 {
 506         xdr_reserve_space(xdr, 0);
 507 }
 508
 509 /*
 510  * 20.1.  Operation 3: CB_GETATTR - Get Attributes
 511  */
 512 static void nfs4_xdr_enc_cb_getattr(struct rpc_rqst *req,
 513                 struct xdr_stream *xdr, const void *data)
 514 {
 515         const struct nfsd4_callback *cb = data;
 516         struct nfs4_cb_fattr *ncf =
 517                 container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
 518         struct nfs4_cb_compound_hdr hdr = {
 519                 .ident = cb->cb_clp->cl_cb_ident,
 520                 .minorversion = cb->cb_clp->cl_minorversion,
 521         };
 522
 523         encode_cb_compound4args(xdr, &hdr);
 524         encode_cb_sequence4args(xdr, cb, &hdr);
 525         encode_cb_getattr4args(xdr, &hdr, ncf);
 526         encode_cb_nops(&hdr);
 527 }
 528
 529 /*
 530  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
 531  */
 532 static void nfs4_xdr_enc_cb_recall(struct rpc_rqst *req, struct xdr_stream *xdr,
 533                                    const void *data)
 534 {
 535         const struct nfsd4_callback *cb = data;
 536         const struct nfs4_delegation *dp = cb_to_delegation(cb);
 537         struct nfs4_cb_compound_hdr hdr = {
 538                 .ident = cb->cb_clp->cl_cb_ident,
 539                 .minorversion = cb->cb_clp->cl_minorversion,
 540         };
 541
 542         encode_cb_compound4args(xdr, &hdr);
 543         encode_cb_sequence4args(xdr, cb, &hdr);
 544         encode_cb_recall4args(xdr, dp, &hdr);
 545         encode_cb_nops(&hdr);
 546 }
 547
 548 /*
 549  * 20.6. Operation 8: CB_RECALL_ANY - Keep Any N Recallable Objects
 550  */
 551 static void
 552 nfs4_xdr_enc_cb_recall_any(struct rpc_rqst *req,
 553                 struct xdr_stream *xdr, const void *data)
 554 {
 555         const struct nfsd4_callback *cb = data;
 556         struct nfsd4_cb_recall_any *ra;
 557         struct nfs4_cb_compound_hdr hdr = {
 558                 .ident = cb->cb_clp->cl_cb_ident,
 559                 .minorversion = cb->cb_clp->cl_minorversion,
 560         };
 561
 562         ra = container_of(cb, struct nfsd4_cb_recall_any, ra_cb);
 563         encode_cb_compound4args(xdr, &hdr);
 564         encode_cb_sequence4args(xdr, cb, &hdr);
 565         encode_cb_recallany4args(xdr, &hdr, ra);
 566         encode_cb_nops(&hdr);
 567 }
 568
 569 /*
 570  * NFSv4.0 and NFSv4.1 XDR decode functions
 571  *
 572  * NFSv4.0 callback result types are defined in section 15 of RFC
 573  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
 574  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
 575  * Protocol".
 576  */
 577
 578 static int nfs4_xdr_dec_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
 579                                 void *__unused)
 580 {
 581         return 0;
 582 }
 583
 584 /*
 585  * 20.1.  Operation 3: CB_GETATTR - Get Attributes
 586  */
 587 static int nfs4_xdr_dec_cb_getattr(struct rpc_rqst *rqstp,
 588                                   struct xdr_stream *xdr,
 589                                   void *data)
 590 {
 591         struct nfsd4_callback *cb = data;
 592         struct nfs4_cb_compound_hdr hdr;
 593         int status;
 594         u32 bitmap[3] = {0};
 595         u32 attrlen;
 596         struct nfs4_cb_fattr *ncf =
 597                 container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
 598
 599         status = decode_cb_compound4res(xdr, &hdr);
 600         if (unlikely(status))
 601                 return status;
 602
 603         status = decode_cb_sequence4res(xdr, cb);
 604         if (unlikely(status || cb->cb_seq_status))
 605                 return status;
 606
 607         status = decode_cb_op_status(xdr, OP_CB_GETATTR, &cb->cb_status);
 608         if (status)
 609                 return status;
 610         if (xdr_stream_decode_uint32_array(xdr, bitmap, 3) < 0)
 611                 return -NFSERR_BAD_XDR;
 612         if (xdr_stream_decode_u32(xdr, &attrlen) < 0)
 613                 return -NFSERR_BAD_XDR;
 614         if (attrlen > (sizeof(ncf->ncf_cb_change) + sizeof(ncf->ncf_cb_fsize)))
 615                 return -NFSERR_BAD_XDR;
 616         status = decode_cb_fattr4(xdr, bitmap, ncf);
 617         return status;
 618 }
 619
 620 /*
 621  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
 622  */
 623 static int nfs4_xdr_dec_cb_recall(struct rpc_rqst *rqstp,
 624                                   struct xdr_stream *xdr,
 625                                   void *data)
 626 {
 627         struct nfsd4_callback *cb = data;
 628         struct nfs4_cb_compound_hdr hdr;
 629         int status;
 630
 631         status = decode_cb_compound4res(xdr, &hdr);
 632         if (unlikely(status))
 633                 return status;
 634
 635         status = decode_cb_sequence4res(xdr, cb);
 636         if (unlikely(status || cb->cb_seq_status))
 637                 return status;
 638
 639         return decode_cb_op_status(xdr, OP_CB_RECALL, &cb->cb_status);
 640 }
 641
 642 /*
 643  * 20.6. Operation 8: CB_RECALL_ANY - Keep Any N Recallable Objects
 644  */
 645 static int
 646 nfs4_xdr_dec_cb_recall_any(struct rpc_rqst *rqstp,
 647                                   struct xdr_stream *xdr,
 648                                   void *data)
 649 {
 650         struct nfsd4_callback *cb = data;
 651         struct nfs4_cb_compound_hdr hdr;
 652         int status;
 653
 654         status = decode_cb_compound4res(xdr, &hdr);
 655         if (unlikely(status))
 656                 return status;
 657         status = decode_cb_sequence4res(xdr, cb);
 658         if (unlikely(status || cb->cb_seq_status))
 659                 return status;
 660         status =  decode_cb_op_status(xdr, OP_CB_RECALL_ANY, &cb->cb_status);
 661         return status;
 662 }
 663
 664 #ifdef CONFIG_NFSD_PNFS
 665 /*
 666  * CB_LAYOUTRECALL4args
 667  *
 668  *      struct layoutrecall_file4 {
 669  *              nfs_fh4         lor_fh;
 670  *              offset4         lor_offset;
 671  *              length4         lor_length;
 672  *              stateid4        lor_stateid;
 673  *      };
 674  *
 675  *      union layoutrecall4 switch(layoutrecall_type4 lor_recalltype) {
 676  *      case LAYOUTRECALL4_FILE:
 677  *              layoutrecall_file4 lor_layout;
 678  *      case LAYOUTRECALL4_FSID:
 679  *              fsid4              lor_fsid;
 680  *      case LAYOUTRECALL4_ALL:
 681  *              void;
 682  *      };
 683  *
 684  *      struct CB_LAYOUTRECALL4args {
 685  *              layouttype4             clora_type;
 686  *              layoutiomode4           clora_iomode;
 687  *              bool                    clora_changed;
 688  *              layoutrecall4           clora_recall;
 689  *      };
 690  */
 691 static void encode_cb_layout4args(struct xdr_stream *xdr,
 692                                   const struct nfs4_layout_stateid *ls,
 693                                   struct nfs4_cb_compound_hdr *hdr)
 694 {
 695         __be32 *p;
 696
 697         BUG_ON(hdr->minorversion == 0);
 698
 699         p = xdr_reserve_space(xdr, 5 * 4);
 700         *p++ = cpu_to_be32(OP_CB_LAYOUTRECALL);
 701         *p++ = cpu_to_be32(ls->ls_layout_type);
 702         *p++ = cpu_to_be32(IOMODE_ANY);
 703         *p++ = cpu_to_be32(1);
 704         *p = cpu_to_be32(RETURN_FILE);
 705
 706         encode_nfs_fh4(xdr, &ls->ls_stid.sc_file->fi_fhandle);
 707
 708         p = xdr_reserve_space(xdr, 2 * 8);
 709         p = xdr_encode_hyper(p, 0);
 710         xdr_encode_hyper(p, NFS4_MAX_UINT64);
 711
 712         encode_stateid4(xdr, &ls->ls_recall_sid);
 713
 714         hdr->nops++;
 715 }
 716
 717 static void nfs4_xdr_enc_cb_layout(struct rpc_rqst *req,
 718                                    struct xdr_stream *xdr,
 719                                    const void *data)
 720 {
 721         const struct nfsd4_callback *cb = data;
 722         const struct nfs4_layout_stateid *ls =
 723                 container_of(cb, struct nfs4_layout_stateid, ls_recall);
 724         struct nfs4_cb_compound_hdr hdr = {
 725                 .ident = 0,
 726                 .minorversion = cb->cb_clp->cl_minorversion,
 727         };
 728
 729         encode_cb_compound4args(xdr, &hdr);
 730         encode_cb_sequence4args(xdr, cb, &hdr);
 731         encode_cb_layout4args(xdr, ls, &hdr);
 732         encode_cb_nops(&hdr);
 733 }
 734
 735 static int nfs4_xdr_dec_cb_layout(struct rpc_rqst *rqstp,
 736                                   struct xdr_stream *xdr,
 737                                   void *data)
 738 {
 739         struct nfsd4_callback *cb = data;
 740         struct nfs4_cb_compound_hdr hdr;
 741         int status;
 742
 743         status = decode_cb_compound4res(xdr, &hdr);
 744         if (unlikely(status))
 745                 return status;
 746
 747         status = decode_cb_sequence4res(xdr, cb);
 748         if (unlikely(status || cb->cb_seq_status))
 749                 return status;
 750
 751         return decode_cb_op_status(xdr, OP_CB_LAYOUTRECALL, &cb->cb_status);
 752 }
 753 #endif /* CONFIG_NFSD_PNFS */
 754
 755 static void encode_stateowner(struct xdr_stream *xdr, struct nfs4_stateowner *so)
 756 {
 757         __be32  *p;
 758
 759         p = xdr_reserve_space(xdr, 8 + 4 + so->so_owner.len);
 760         p = xdr_encode_opaque_fixed(p, &so->so_client->cl_clientid, 8);
 761         xdr_encode_opaque(p, so->so_owner.data, so->so_owner.len);
 762 }
 763
 764 static void nfs4_xdr_enc_cb_notify_lock(struct rpc_rqst *req,
 765                                         struct xdr_stream *xdr,
 766                                         const void *data)
 767 {
 768         const struct nfsd4_callback *cb = data;
 769         const struct nfsd4_blocked_lock *nbl =
 770                 container_of(cb, struct nfsd4_blocked_lock, nbl_cb);
 771         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)nbl->nbl_lock.c.flc_owner;
 772         struct nfs4_cb_compound_hdr hdr = {
 773                 .ident = 0,
 774                 .minorversion = cb->cb_clp->cl_minorversion,
 775         };
 776
 777         __be32 *p;
 778
 779         BUG_ON(hdr.minorversion == 0);
 780
 781         encode_cb_compound4args(xdr, &hdr);
 782         encode_cb_sequence4args(xdr, cb, &hdr);
 783
 784         p = xdr_reserve_space(xdr, 4);
 785         *p = cpu_to_be32(OP_CB_NOTIFY_LOCK);
 786         encode_nfs_fh4(xdr, &nbl->nbl_fh);
 787         encode_stateowner(xdr, &lo->lo_owner);
 788         hdr.nops++;
 789
 790         encode_cb_nops(&hdr);
 791 }
 792
 793 static int nfs4_xdr_dec_cb_notify_lock(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_NOTIFY_LOCK, &cb->cb_status);
 810 }
 811
 812 /*
 813  * struct write_response4 {
 814  *      stateid4        wr_callback_id<1>;
 815  *      length4         wr_count;
 816  *      stable_how4     wr_committed;
 817  *      verifier4       wr_writeverf;
 818  * };
 819  * union offload_info4 switch (nfsstat4 coa_status) {
 820  *      case NFS4_OK:
 821  *              write_response4 coa_resok4;
 822  *      default:
 823  *              length4         coa_bytes_copied;
 824  * };
 825  * struct CB_OFFLOAD4args {
 826  *      nfs_fh4         coa_fh;
 827  *      stateid4        coa_stateid;
 828  *      offload_info4   coa_offload_info;
 829  * };
 830  */
 831 static void encode_offload_info4(struct xdr_stream *xdr,
 832                                  const struct nfsd4_cb_offload *cbo)
 833 {
 834         __be32 *p;
 835
 836         p = xdr_reserve_space(xdr, 4);
 837         *p = cbo->co_nfserr;
 838         switch (cbo->co_nfserr) {
 839         case nfs_ok:
 840                 p = xdr_reserve_space(xdr, 4 + 8 + 4 + NFS4_VERIFIER_SIZE);
 841                 p = xdr_encode_empty_array(p);
 842                 p = xdr_encode_hyper(p, cbo->co_res.wr_bytes_written);
 843                 *p++ = cpu_to_be32(cbo->co_res.wr_stable_how);
 844                 p = xdr_encode_opaque_fixed(p, cbo->co_res.wr_verifier.data,
 845                                             NFS4_VERIFIER_SIZE);
 846                 break;
 847         default:
 848                 p = xdr_reserve_space(xdr, 8);
 849                 /* We always return success if bytes were written */
 850                 p = xdr_encode_hyper(p, 0);
 851         }
 852 }
 853
 854 static void encode_cb_offload4args(struct xdr_stream *xdr,
 855                                    const struct nfsd4_cb_offload *cbo,
 856                                    struct nfs4_cb_compound_hdr *hdr)
 857 {
 858         __be32 *p;
 859
 860         p = xdr_reserve_space(xdr, 4);
 861         *p = cpu_to_be32(OP_CB_OFFLOAD);
 862         encode_nfs_fh4(xdr, &cbo->co_fh);
 863         encode_stateid4(xdr, &cbo->co_res.cb_stateid);
 864         encode_offload_info4(xdr, cbo);
 865
 866         hdr->nops++;
 867 }
 868
 869 static void nfs4_xdr_enc_cb_offload(struct rpc_rqst *req,
 870                                     struct xdr_stream *xdr,
 871                                     const void *data)
 872 {
 873         const struct nfsd4_callback *cb = data;
 874         const struct nfsd4_cb_offload *cbo =
 875                 container_of(cb, struct nfsd4_cb_offload, co_cb);
 876         struct nfs4_cb_compound_hdr hdr = {
 877                 .ident = 0,
 878                 .minorversion = cb->cb_clp->cl_minorversion,
 879         };
 880
 881         encode_cb_compound4args(xdr, &hdr);
 882         encode_cb_sequence4args(xdr, cb, &hdr);
 883         encode_cb_offload4args(xdr, cbo, &hdr);
 884         encode_cb_nops(&hdr);
 885 }
 886
 887 static int nfs4_xdr_dec_cb_offload(struct rpc_rqst *rqstp,
 888                                    struct xdr_stream *xdr,
 889                                    void *data)
 890 {
 891         struct nfsd4_callback *cb = data;
 892         struct nfs4_cb_compound_hdr hdr;
 893         int status;
 894
 895         status = decode_cb_compound4res(xdr, &hdr);
 896         if (unlikely(status))
 897                 return status;
 898
 899         status = decode_cb_sequence4res(xdr, cb);
 900         if (unlikely(status || cb->cb_seq_status))
 901                 return status;
 902
 903         return decode_cb_op_status(xdr, OP_CB_OFFLOAD, &cb->cb_status);
 904 }
 905 /*
 906  * RPC procedure tables
 907  */
 908 #define PROC(proc, call, argtype, restype)                              \
 909 [NFSPROC4_CLNT_##proc] = {                                              \
 910         .p_proc    = NFSPROC4_CB_##call,                                \
 911         .p_encode  = nfs4_xdr_enc_##argtype,            \
 912         .p_decode  = nfs4_xdr_dec_##restype,                            \
 913         .p_arglen  = NFS4_enc_##argtype##_sz,                           \
 914         .p_replen  = NFS4_dec_##restype##_sz,                           \
 915         .p_statidx = NFSPROC4_CB_##call,                                \
 916         .p_name    = #proc,                                             \
 917 }
 918
 919 static const struct rpc_procinfo nfs4_cb_procedures[] = {
 920         PROC(CB_NULL,   NULL,           cb_null,        cb_null),
 921         PROC(CB_RECALL, COMPOUND,       cb_recall,      cb_recall),
 922 #ifdef CONFIG_NFSD_PNFS
 923         PROC(CB_LAYOUT, COMPOUND,       cb_layout,      cb_layout),
 924 #endif
 925         PROC(CB_NOTIFY_LOCK,    COMPOUND,       cb_notify_lock, cb_notify_lock),
 926         PROC(CB_OFFLOAD,        COMPOUND,       cb_offload,     cb_offload),
 927         PROC(CB_RECALL_ANY,     COMPOUND,       cb_recall_any,  cb_recall_any),
 928         PROC(CB_GETATTR,        COMPOUND,       cb_getattr,     cb_getattr),
 929 };
 930
 931 static unsigned int nfs4_cb_counts[ARRAY_SIZE(nfs4_cb_procedures)];
 932 static const struct rpc_version nfs_cb_version4 = {
 933 /*
 934  * Note on the callback rpc program version number: despite language in rfc
 935  * 5661 section 18.36.3 requiring servers to use 4 in this field, the
 936  * official xdr descriptions for both 4.0 and 4.1 specify version 1, and
 937  * in practice that appears to be what implementations use.  The section
 938  * 18.36.3 language is expected to be fixed in an erratum.
 939  */
 940         .number                 = 1,
 941         .nrprocs                = ARRAY_SIZE(nfs4_cb_procedures),
 942         .procs                  = nfs4_cb_procedures,
 943         .counts                 = nfs4_cb_counts,
 944 };
 945
 946 static const struct rpc_version *nfs_cb_version[2] = {
 947         [1] = &nfs_cb_version4,
 948 };
 949
 950 static const struct rpc_program cb_program;
 951
 952 static struct rpc_stat cb_stats = {
 953         .program                = &cb_program
 954 };
 955
 956 #define NFS4_CALLBACK 0x40000000
 957 static const struct rpc_program cb_program = {
 958         .name                   = "nfs4_cb",
 959         .number                 = NFS4_CALLBACK,
 960         .nrvers                 = ARRAY_SIZE(nfs_cb_version),
 961         .version                = nfs_cb_version,
 962         .stats                  = &cb_stats,
 963         .pipe_dir_name          = "nfsd4_cb",
 964 };
 965
 966 static int max_cb_time(struct net *net)
 967 {
 968         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 969
 970         /*
 971          * nfsd4_lease is set to at most one hour in __nfsd4_write_time,
 972          * so we can use 32-bit math on it. Warn if that assumption
 973          * ever stops being true.
 974          */
 975         if (WARN_ON_ONCE(nn->nfsd4_lease > 3600))
 976                 return 360 * HZ;
 977
 978         return max(((u32)nn->nfsd4_lease)/10, 1u) * HZ;
 979 }
 980
 981 static struct workqueue_struct *callback_wq;
 982
 983 static bool nfsd4_queue_cb(struct nfsd4_callback *cb)
 984 {
 985         trace_nfsd_cb_queue(cb->cb_clp, cb);
 986         return queue_delayed_work(callback_wq, &cb->cb_work, 0);
 987 }
 988
 989 static void nfsd4_queue_cb_delayed(struct nfsd4_callback *cb,
 990                                    unsigned long msecs)
 991 {
 992         trace_nfsd_cb_queue(cb->cb_clp, cb);
 993         queue_delayed_work(callback_wq, &cb->cb_work,
 994                            msecs_to_jiffies(msecs));
 995 }
 996
 997 static void nfsd41_cb_inflight_begin(struct nfs4_client *clp)
 998 {
 999         atomic_inc(&clp->cl_cb_inflight);
1000 }
1001
1002 static void nfsd41_cb_inflight_end(struct nfs4_client *clp)
1003 {
1004
1005         if (atomic_dec_and_test(&clp->cl_cb_inflight))
1006                 wake_up_var(&clp->cl_cb_inflight);
1007 }
1008
1009 static void nfsd41_cb_inflight_wait_complete(struct nfs4_client *clp)
1010 {
1011         wait_var_event(&clp->cl_cb_inflight,
1012                         !atomic_read(&clp->cl_cb_inflight));
1013 }
1014
1015 static const struct cred *get_backchannel_cred(struct nfs4_client *clp, struct rpc_clnt *client, struct nfsd4_session *ses)
1016 {
1017         if (clp->cl_minorversion == 0) {
1018                 client->cl_principal = clp->cl_cred.cr_targ_princ ?
1019                         clp->cl_cred.cr_targ_princ : "nfs";
1020
1021                 return get_cred(rpc_machine_cred());
1022         } else {
1023                 struct cred *kcred;
1024
1025                 kcred = prepare_kernel_cred(&init_task);
1026                 if (!kcred)
1027                         return NULL;
1028
1029                 kcred->fsuid = ses->se_cb_sec.uid;
1030                 kcred->fsgid = ses->se_cb_sec.gid;
1031                 return kcred;
1032         }
1033 }
1034
1035 static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn, struct nfsd4_session *ses)
1036 {
1037         int maxtime = max_cb_time(clp->net);
1038         struct rpc_timeout      timeparms = {
1039                 .to_initval     = maxtime,
1040                 .to_retries     = 0,
1041                 .to_maxval      = maxtime,
1042         };
1043         struct rpc_create_args args = {
1044                 .net            = clp->net,
1045                 .address        = (struct sockaddr *) &conn->cb_addr,
1046                 .addrsize       = conn->cb_addrlen,
1047                 .saddress       = (struct sockaddr *) &conn->cb_saddr,
1048                 .timeout        = &timeparms,
1049                 .program        = &cb_program,
1050                 .version        = 1,
1051                 .flags          = (RPC_CLNT_CREATE_NOPING | RPC_CLNT_CREATE_QUIET),
1052                 .cred           = current_cred(),
1053         };
1054         struct rpc_clnt *client;
1055         const struct cred *cred;
1056
1057         if (clp->cl_minorversion == 0) {
1058                 if (!clp->cl_cred.cr_principal &&
1059                     (clp->cl_cred.cr_flavor >= RPC_AUTH_GSS_KRB5)) {
1060                         trace_nfsd_cb_setup_err(clp, -EINVAL);
1061                         return -EINVAL;
1062                 }
1063                 args.client_name = clp->cl_cred.cr_principal;
1064                 args.prognumber = conn->cb_prog;
1065                 args.protocol = XPRT_TRANSPORT_TCP;
1066                 args.authflavor = clp->cl_cred.cr_flavor;
1067                 clp->cl_cb_ident = conn->cb_ident;
1068         } else {
1069                 if (!conn->cb_xprt)
1070                         return -EINVAL;
1071                 clp->cl_cb_session = ses;
1072                 args.bc_xprt = conn->cb_xprt;
1073                 args.prognumber = clp->cl_cb_session->se_cb_prog;
1074                 args.protocol = conn->cb_xprt->xpt_class->xcl_ident |
1075                                 XPRT_TRANSPORT_BC;
1076                 args.authflavor = ses->se_cb_sec.flavor;
1077         }
1078         /* Create RPC client */
1079         client = rpc_create(&args);
1080         if (IS_ERR(client)) {
1081                 trace_nfsd_cb_setup_err(clp, PTR_ERR(client));
1082                 return PTR_ERR(client);
1083         }
1084         cred = get_backchannel_cred(clp, client, ses);
1085         if (!cred) {
1086                 trace_nfsd_cb_setup_err(clp, -ENOMEM);
1087                 rpc_shutdown_client(client);
1088                 return -ENOMEM;
1089         }
1090
1091         if (clp->cl_minorversion != 0)
1092                 clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
1093         clp->cl_cb_client = client;
1094         clp->cl_cb_cred = cred;
1095         rcu_read_lock();
1096         trace_nfsd_cb_setup(clp, rpc_peeraddr2str(client, RPC_DISPLAY_NETID),
1097                             args.authflavor);
1098         rcu_read_unlock();
1099         return 0;
1100 }
1101
1102 static void nfsd4_mark_cb_state(struct nfs4_client *clp, int newstate)
1103 {
1104         if (clp->cl_cb_state != newstate) {
1105                 clp->cl_cb_state = newstate;
1106                 trace_nfsd_cb_new_state(clp);
1107         }
1108 }
1109
1110 static void nfsd4_mark_cb_down(struct nfs4_client *clp)
1111 {
1112         if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
1113                 return;
1114         nfsd4_mark_cb_state(clp, NFSD4_CB_DOWN);
1115 }
1116
1117 static void nfsd4_mark_cb_fault(struct nfs4_client *clp)
1118 {
1119         if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
1120                 return;
1121         nfsd4_mark_cb_state(clp, NFSD4_CB_FAULT);
1122 }
1123
1124 static void nfsd4_cb_probe_done(struct rpc_task *task, void *calldata)
1125 {
1126         struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
1127
1128         if (task->tk_status)
1129                 nfsd4_mark_cb_down(clp);
1130         else
1131                 nfsd4_mark_cb_state(clp, NFSD4_CB_UP);
1132 }
1133
1134 static void nfsd4_cb_probe_release(void *calldata)
1135 {
1136         struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
1137
1138         nfsd41_cb_inflight_end(clp);
1139
1140 }
1141
1142 static const struct rpc_call_ops nfsd4_cb_probe_ops = {
1143         /* XXX: release method to ensure we set the cb channel down if
1144          * necessary on early failure? */
1145         .rpc_call_done = nfsd4_cb_probe_done,
1146         .rpc_release = nfsd4_cb_probe_release,
1147 };
1148
1149 /*
1150  * Poke the callback thread to process any updates to the callback
1151  * parameters, and send a null probe.
1152  */
1153 void nfsd4_probe_callback(struct nfs4_client *clp)
1154 {
1155         trace_nfsd_cb_probe(clp);
1156         nfsd4_mark_cb_state(clp, NFSD4_CB_UNKNOWN);
1157         set_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
1158         nfsd4_run_cb(&clp->cl_cb_null);
1159 }
1160
1161 void nfsd4_probe_callback_sync(struct nfs4_client *clp)
1162 {
1163         nfsd4_probe_callback(clp);
1164         flush_workqueue(callback_wq);
1165 }
1166
1167 void nfsd4_change_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
1168 {
1169         nfsd4_mark_cb_state(clp, NFSD4_CB_UNKNOWN);
1170         spin_lock(&clp->cl_lock);
1171         memcpy(&clp->cl_cb_conn, conn, sizeof(struct nfs4_cb_conn));
1172         spin_unlock(&clp->cl_lock);
1173 }
1174
1175 /*
1176  * There's currently a single callback channel slot.
1177  * If the slot is available, then mark it busy.  Otherwise, set the
1178  * thread for sleeping on the callback RPC wait queue.
1179  */
1180 static bool nfsd41_cb_get_slot(struct nfsd4_callback *cb, struct rpc_task *task)
1181 {
1182         struct nfs4_client *clp = cb->cb_clp;
1183
1184         if (!cb->cb_holds_slot &&
1185             test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
1186                 rpc_sleep_on(&clp->cl_cb_waitq, task, NULL);
1187                 /* Race breaker */
1188                 if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
1189                         dprintk("%s slot is busy\n", __func__);
1190                         return false;
1191                 }
1192                 rpc_wake_up_queued_task(&clp->cl_cb_waitq, task);
1193         }
1194         cb->cb_holds_slot = true;
1195         return true;
1196 }
1197
1198 static void nfsd41_cb_release_slot(struct nfsd4_callback *cb)
1199 {
1200         struct nfs4_client *clp = cb->cb_clp;
1201
1202         if (cb->cb_holds_slot) {
1203                 cb->cb_holds_slot = false;
1204                 clear_bit(0, &clp->cl_cb_slot_busy);
1205                 rpc_wake_up_next(&clp->cl_cb_waitq);
1206         }
1207 }
1208
1209 static void nfsd41_destroy_cb(struct nfsd4_callback *cb)
1210 {
1211         struct nfs4_client *clp = cb->cb_clp;
1212
1213         trace_nfsd_cb_destroy(clp, cb);
1214         nfsd41_cb_release_slot(cb);
1215         if (cb->cb_ops && cb->cb_ops->release)
1216                 cb->cb_ops->release(cb);
1217         nfsd41_cb_inflight_end(clp);
1218 }
1219
1220 /*
1221  * TODO: cb_sequence should support referring call lists, cachethis, multiple
1222  * slots, and mark callback channel down on communication errors.
1223  */
1224 static void nfsd4_cb_prepare(struct rpc_task *task, void *calldata)
1225 {
1226         struct nfsd4_callback *cb = calldata;
1227         struct nfs4_client *clp = cb->cb_clp;
1228         u32 minorversion = clp->cl_minorversion;
1229
1230         /*
1231          * cb_seq_status is only set in decode_cb_sequence4res,
1232          * and so will remain 1 if an rpc level failure occurs.
1233          */
1234         cb->cb_seq_status = 1;
1235         cb->cb_status = 0;
1236         if (minorversion && !nfsd41_cb_get_slot(cb, task))
1237                 return;
1238         rpc_call_start(task);
1239 }
1240
1241 static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback *cb)
1242 {
1243         struct nfs4_client *clp = cb->cb_clp;
1244         struct nfsd4_session *session = clp->cl_cb_session;
1245         bool ret = true;
1246
1247         if (!clp->cl_minorversion) {
1248                 /*
1249                  * If the backchannel connection was shut down while this
1250                  * task was queued, we need to resubmit it after setting up
1251                  * a new backchannel connection.
1252                  *
1253                  * Note that if we lost our callback connection permanently
1254                  * the submission code will error out, so we don't need to
1255                  * handle that case here.
1256                  */
1257                 if (RPC_SIGNALLED(task))
1258                         goto need_restart;
1259
1260                 return true;
1261         }
1262
1263         if (!cb->cb_holds_slot)
1264                 goto need_restart;
1265
1266         /* This is the operation status code for CB_SEQUENCE */
1267         trace_nfsd_cb_seq_status(task, cb);
1268         switch (cb->cb_seq_status) {
1269         case 0:
1270                 /*
1271                  * No need for lock, access serialized in nfsd4_cb_prepare
1272                  *
1273                  * RFC5661 20.9.3
1274                  * If CB_SEQUENCE returns an error, then the state of the slot
1275                  * (sequence ID, cached reply) MUST NOT change.
1276                  */
1277                 ++session->se_cb_seq_nr;
1278                 break;
1279         case -ESERVERFAULT:
1280                 ++session->se_cb_seq_nr;
1281                 nfsd4_mark_cb_fault(cb->cb_clp);
1282                 ret = false;
1283                 break;
1284         case 1:
1285                 /*
1286                  * cb_seq_status remains 1 if an RPC Reply was never
1287                  * received. NFSD can't know if the client processed
1288                  * the CB_SEQUENCE operation. Ask the client to send a
1289                  * DESTROY_SESSION to recover.
1290                  */
1291                 fallthrough;
1292         case -NFS4ERR_BADSESSION:
1293                 nfsd4_mark_cb_fault(cb->cb_clp);
1294                 ret = false;
1295                 goto need_restart;
1296         case -NFS4ERR_DELAY:
1297                 cb->cb_seq_status = 1;
1298                 if (!rpc_restart_call(task))
1299                         goto out;
1300
1301                 rpc_delay(task, 2 * HZ);
1302                 return false;
1303         case -NFS4ERR_BADSLOT:
1304                 goto retry_nowait;
1305         case -NFS4ERR_SEQ_MISORDERED:
1306                 if (session->se_cb_seq_nr != 1) {
1307                         session->se_cb_seq_nr = 1;
1308                         goto retry_nowait;
1309                 }
1310                 break;
1311         default:
1312                 nfsd4_mark_cb_fault(cb->cb_clp);
1313         }
1314         nfsd41_cb_release_slot(cb);
1315
1316         trace_nfsd_cb_free_slot(task, cb);
1317
1318         if (RPC_SIGNALLED(task))
1319                 goto need_restart;
1320 out:
1321         return ret;
1322 retry_nowait:
1323         if (rpc_restart_call_prepare(task))
1324                 ret = false;
1325         goto out;
1326 need_restart:
1327         if (!test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags)) {
1328                 trace_nfsd_cb_restart(clp, cb);
1329                 task->tk_status = 0;
1330                 cb->cb_need_restart = true;
1331         }
1332         return false;
1333 }
1334
1335 static void nfsd4_cb_done(struct rpc_task *task, void *calldata)
1336 {
1337         struct nfsd4_callback *cb = calldata;
1338         struct nfs4_client *clp = cb->cb_clp;
1339
1340         if (!nfsd4_cb_sequence_done(task, cb))
1341                 return;
1342
1343         if (cb->cb_status) {
1344                 WARN_ON_ONCE(task->tk_status);
1345                 task->tk_status = cb->cb_status;
1346         }
1347
1348         switch (cb->cb_ops->done(cb, task)) {
1349         case 0:
1350                 task->tk_status = 0;
1351                 rpc_restart_call_prepare(task);
1352                 return;
1353         case 1:
1354                 switch (task->tk_status) {
1355                 case -EIO:
1356                 case -ETIMEDOUT:
1357                 case -EACCES:
1358                         nfsd4_mark_cb_down(clp);
1359                 }
1360                 break;
1361         default:
1362                 BUG();
1363         }
1364 }
1365
1366 static void nfsd4_cb_release(void *calldata)
1367 {
1368         struct nfsd4_callback *cb = calldata;
1369
1370         if (cb->cb_need_restart)
1371                 nfsd4_queue_cb(cb);
1372         else
1373                 nfsd41_destroy_cb(cb);
1374
1375 }
1376
1377 static const struct rpc_call_ops nfsd4_cb_ops = {
1378         .rpc_call_prepare = nfsd4_cb_prepare,
1379         .rpc_call_done = nfsd4_cb_done,
1380         .rpc_release = nfsd4_cb_release,
1381 };
1382
1383 int nfsd4_create_callback_queue(void)
1384 {
1385         callback_wq = alloc_ordered_workqueue("nfsd4_callbacks", 0);
1386         if (!callback_wq)
1387                 return -ENOMEM;
1388         return 0;
1389 }
1390
1391 void nfsd4_destroy_callback_queue(void)
1392 {
1393         destroy_workqueue(callback_wq);
1394 }
1395
1396 /* must be called under the state lock */
1397 void nfsd4_shutdown_callback(struct nfs4_client *clp)
1398 {
1399         if (clp->cl_cb_state != NFSD4_CB_UNKNOWN)
1400                 trace_nfsd_cb_shutdown(clp);
1401
1402         set_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags);
1403         /*
1404          * Note this won't actually result in a null callback;
1405          * instead, nfsd4_run_cb_null() will detect the killed
1406          * client, destroy the rpc client, and stop:
1407          */
1408         nfsd4_run_cb(&clp->cl_cb_null);
1409         flush_workqueue(callback_wq);
1410         nfsd41_cb_inflight_wait_complete(clp);
1411 }
1412
1413 static struct nfsd4_conn * __nfsd4_find_backchannel(struct nfs4_client *clp)
1414 {
1415         struct nfsd4_session *s;
1416         struct nfsd4_conn *c;
1417
1418         lockdep_assert_held(&clp->cl_lock);
1419
1420         list_for_each_entry(s, &clp->cl_sessions, se_perclnt) {
1421                 list_for_each_entry(c, &s->se_conns, cn_persession) {
1422                         if (c->cn_flags & NFS4_CDFC4_BACK)
1423                                 return c;
1424                 }
1425         }
1426         return NULL;
1427 }
1428
1429 /*
1430  * Note there isn't a lot of locking in this code; instead we depend on
1431  * the fact that it is run from the callback_wq, which won't run two
1432  * work items at once.  So, for example, callback_wq handles all access
1433  * of cl_cb_client and all calls to rpc_create or rpc_shutdown_client.
1434  */
1435 static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
1436 {
1437         struct nfs4_cb_conn conn;
1438         struct nfs4_client *clp = cb->cb_clp;
1439         struct nfsd4_session *ses = NULL;
1440         struct nfsd4_conn *c;
1441         int err;
1442
1443         trace_nfsd_cb_bc_update(clp, cb);
1444
1445         /*
1446          * This is either an update, or the client dying; in either case,
1447          * kill the old client:
1448          */
1449         if (clp->cl_cb_client) {
1450                 trace_nfsd_cb_bc_shutdown(clp, cb);
1451                 rpc_shutdown_client(clp->cl_cb_client);
1452                 clp->cl_cb_client = NULL;
1453                 put_cred(clp->cl_cb_cred);
1454                 clp->cl_cb_cred = NULL;
1455         }
1456         if (clp->cl_cb_conn.cb_xprt) {
1457                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1458                 clp->cl_cb_conn.cb_xprt = NULL;
1459         }
1460         if (test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags))
1461                 return;
1462
1463         spin_lock(&clp->cl_lock);
1464         /*
1465          * Only serialized callback code is allowed to clear these
1466          * flags; main nfsd code can only set them:
1467          */
1468         WARN_ON(!(clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK));
1469         clear_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
1470
1471         memcpy(&conn, &cb->cb_clp->cl_cb_conn, sizeof(struct nfs4_cb_conn));
1472         c = __nfsd4_find_backchannel(clp);
1473         if (c) {
1474                 svc_xprt_get(c->cn_xprt);
1475                 conn.cb_xprt = c->cn_xprt;
1476                 ses = c->cn_session;
1477         }
1478         spin_unlock(&clp->cl_lock);
1479
1480         err = setup_callback_client(clp, &conn, ses);
1481         if (err) {
1482                 nfsd4_mark_cb_down(clp);
1483                 if (c)
1484                         svc_xprt_put(c->cn_xprt);
1485                 return;
1486         }
1487 }
1488
1489 static void
1490 nfsd4_run_cb_work(struct work_struct *work)
1491 {
1492         struct nfsd4_callback *cb =
1493                 container_of(work, struct nfsd4_callback, cb_work.work);
1494         struct nfs4_client *clp = cb->cb_clp;
1495         struct rpc_clnt *clnt;
1496         int flags;
1497
1498         trace_nfsd_cb_start(clp);
1499
1500         if (clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK)
1501                 nfsd4_process_cb_update(cb);
1502
1503         clnt = clp->cl_cb_client;
1504         if (!clnt) {
1505                 if (test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags))
1506                         nfsd41_destroy_cb(cb);
1507                 else {
1508                         /*
1509                          * XXX: Ideally, we could wait for the client to
1510                          *      reconnect, but I haven't figured out how
1511                          *      to do that yet.
1512                          */
1513                         nfsd4_queue_cb_delayed(cb, 25);
1514                 }
1515                 return;
1516         }
1517
1518         /*
1519          * Don't send probe messages for 4.1 or later.
1520          */
1521         if (!cb->cb_ops && clp->cl_minorversion) {
1522                 nfsd4_mark_cb_state(clp, NFSD4_CB_UP);
1523                 nfsd41_destroy_cb(cb);
1524                 return;
1525         }
1526
1527         if (cb->cb_need_restart) {
1528                 cb->cb_need_restart = false;
1529         } else {
1530                 if (cb->cb_ops && cb->cb_ops->prepare)
1531                         cb->cb_ops->prepare(cb);
1532         }
1533         cb->cb_msg.rpc_cred = clp->cl_cb_cred;
1534         flags = clp->cl_minorversion ? RPC_TASK_NOCONNECT : RPC_TASK_SOFTCONN;
1535         rpc_call_async(clnt, &cb->cb_msg, RPC_TASK_SOFT | flags,
1536                         cb->cb_ops ? &nfsd4_cb_ops : &nfsd4_cb_probe_ops, cb);
1537 }
1538
1539 void nfsd4_init_cb(struct nfsd4_callback *cb, struct nfs4_client *clp,
1540                 const struct nfsd4_callback_ops *ops, enum nfsd4_cb_op op)
1541 {
1542         cb->cb_clp = clp;
1543         cb->cb_msg.rpc_proc = &nfs4_cb_procedures[op];
1544         cb->cb_msg.rpc_argp = cb;
1545         cb->cb_msg.rpc_resp = cb;
1546         cb->cb_ops = ops;
1547         INIT_DELAYED_WORK(&cb->cb_work, nfsd4_run_cb_work);
1548         cb->cb_status = 0;
1549         cb->cb_need_restart = false;
1550         cb->cb_holds_slot = false;
1551 }
1552
1553 /**
1554  * nfsd4_run_cb - queue up a callback job to run
1555  * @cb: callback to queue
1556  *
1557  * Kick off a callback to do its thing. Returns false if it was already
1558  * on a queue, true otherwise.
1559  */
1560 bool nfsd4_run_cb(struct nfsd4_callback *cb)
1561 {
1562         struct nfs4_client *clp = cb->cb_clp;
1563         bool queued;
1564
1565         nfsd41_cb_inflight_begin(clp);
1566         queued = nfsd4_queue_cb(cb);
1567         if (!queued)
1568                 nfsd41_cb_inflight_end(clp);
1569         return queued;
1570 }