fs/nfsd/nfs4callback.c

   1 /*
   2  *  Copyright (c) 2001 The Regents of the University of Michigan.
   3  *  All rights reserved.
   4  *
   5  *  Kendrick Smith <[email protected]>
   6  *  Andy Adamson <[email protected]>
   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, XDR_UNIT);
 291         if (unlikely(p == NULL))
 292                 goto out_overflow;
 293         hdr->status = be32_to_cpup(p);
 294         /* Ignore the tag */
 295         if (xdr_stream_decode_u32(xdr, &length) < 0)
 296                 goto out_overflow;
 297         if (xdr_inline_decode(xdr, length) == NULL)
 298                 goto out_overflow;
 299         if (xdr_stream_decode_u32(xdr, &hdr->nops) < 0)
 300                 goto out_overflow;
 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         u32 bmap[1];
 368
 369         bmap[0] = FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE;
 370
 371         encode_nfs_cb_opnum4(xdr, OP_CB_GETATTR);
 372         encode_nfs_fh4(xdr, fh);
 373         encode_bitmap4(xdr, bmap, ARRAY_SIZE(bmap));
 374         hdr->nops++;
 375 }
 376
 377 static u32 highest_slotid(struct nfsd4_session *ses)
 378 {
 379         u32 idx;
 380
 381         spin_lock(&ses->se_lock);
 382         idx = fls(~ses->se_cb_slot_avail);
 383         if (idx > 0)
 384                 --idx;
 385         idx = max(idx, ses->se_cb_highest_slot);
 386         spin_unlock(&ses->se_lock);
 387         return idx;
 388 }
 389
 390 /*
 391  * CB_SEQUENCE4args
 392  *
 393  *      struct CB_SEQUENCE4args {
 394  *              sessionid4              csa_sessionid;
 395  *              sequenceid4             csa_sequenceid;
 396  *              slotid4                 csa_slotid;
 397  *              slotid4                 csa_highest_slotid;
 398  *              bool                    csa_cachethis;
 399  *              referring_call_list4    csa_referring_call_lists<>;
 400  *      };
 401  */
 402 static void encode_cb_sequence4args(struct xdr_stream *xdr,
 403                                     const struct nfsd4_callback *cb,
 404                                     struct nfs4_cb_compound_hdr *hdr)
 405 {
 406         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
 407         __be32 *p;
 408
 409         if (hdr->minorversion == 0)
 410                 return;
 411
 412         encode_nfs_cb_opnum4(xdr, OP_CB_SEQUENCE);
 413         encode_sessionid4(xdr, session);
 414
 415         p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4 + 4);
 416         *p++ = cpu_to_be32(session->se_cb_seq_nr[cb->cb_held_slot]);    /* csa_sequenceid */
 417         *p++ = cpu_to_be32(cb->cb_held_slot);           /* csa_slotid */
 418         *p++ = cpu_to_be32(highest_slotid(session)); /* csa_highest_slotid */
 419         *p++ = xdr_zero;                        /* csa_cachethis */
 420         xdr_encode_empty_array(p);              /* csa_referring_call_lists */
 421
 422         hdr->nops++;
 423 }
 424
 425 static void update_cb_slot_table(struct nfsd4_session *ses, u32 target)
 426 {
 427         /* No need to do anything if nothing changed */
 428         if (likely(target == READ_ONCE(ses->se_cb_highest_slot)))
 429                 return;
 430
 431         spin_lock(&ses->se_lock);
 432         if (target > ses->se_cb_highest_slot) {
 433                 int i;
 434
 435                 target = min(target, NFSD_BC_SLOT_TABLE_SIZE - 1);
 436
 437                 /*
 438                  * Growing the slot table. Reset any new sequences to 1.
 439                  *
 440                  * NB: There is some debate about whether the RFC requires this,
 441                  *     but the Linux client expects it.
 442                  */
 443                 for (i = ses->se_cb_highest_slot + 1; i <= target; ++i)
 444                         ses->se_cb_seq_nr[i] = 1;
 445         }
 446         ses->se_cb_highest_slot = target;
 447         spin_unlock(&ses->se_lock);
 448 }
 449
 450 /*
 451  * CB_SEQUENCE4resok
 452  *
 453  *      struct CB_SEQUENCE4resok {
 454  *              sessionid4      csr_sessionid;
 455  *              sequenceid4     csr_sequenceid;
 456  *              slotid4         csr_slotid;
 457  *              slotid4         csr_highest_slotid;
 458  *              slotid4         csr_target_highest_slotid;
 459  *      };
 460  *
 461  *      union CB_SEQUENCE4res switch (nfsstat4 csr_status) {
 462  *      case NFS4_OK:
 463  *              CB_SEQUENCE4resok       csr_resok4;
 464  *      default:
 465  *              void;
 466  *      };
 467  *
 468  * Our current back channel implmentation supports a single backchannel
 469  * with a single slot.
 470  */
 471 static int decode_cb_sequence4resok(struct xdr_stream *xdr,
 472                                     struct nfsd4_callback *cb)
 473 {
 474         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
 475         int status = -ESERVERFAULT;
 476         __be32 *p;
 477         u32 seqid, slotid, target;
 478
 479         /*
 480          * If the server returns different values for sessionID, slotID or
 481          * sequence number, the server is looney tunes.
 482          */
 483         p = xdr_inline_decode(xdr, NFS4_MAX_SESSIONID_LEN + 4 + 4 + 4 + 4);
 484         if (unlikely(p == NULL))
 485                 goto out_overflow;
 486
 487         if (memcmp(p, session->se_sessionid.data, NFS4_MAX_SESSIONID_LEN)) {
 488                 dprintk("NFS: %s Invalid session id\n", __func__);
 489                 goto out;
 490         }
 491         p += XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN);
 492
 493         seqid = be32_to_cpup(p++);
 494         if (seqid != session->se_cb_seq_nr[cb->cb_held_slot]) {
 495                 dprintk("NFS: %s Invalid sequence number\n", __func__);
 496                 goto out;
 497         }
 498
 499         slotid = be32_to_cpup(p++);
 500         if (slotid != cb->cb_held_slot) {
 501                 dprintk("NFS: %s Invalid slotid\n", __func__);
 502                 goto out;
 503         }
 504
 505         p++; // ignore current highest slot value
 506
 507         target = be32_to_cpup(p++);
 508         update_cb_slot_table(session, target);
 509         status = 0;
 510 out:
 511         cb->cb_seq_status = status;
 512         return status;
 513 out_overflow:
 514         status = -EIO;
 515         goto out;
 516 }
 517
 518 static int decode_cb_sequence4res(struct xdr_stream *xdr,
 519                                   struct nfsd4_callback *cb)
 520 {
 521         int status;
 522
 523         if (cb->cb_clp->cl_minorversion == 0)
 524                 return 0;
 525
 526         status = decode_cb_op_status(xdr, OP_CB_SEQUENCE, &cb->cb_seq_status);
 527         if (unlikely(status || cb->cb_seq_status))
 528                 return status;
 529
 530         return decode_cb_sequence4resok(xdr, cb);
 531 }
 532
 533 /*
 534  * NFSv4.0 and NFSv4.1 XDR encode functions
 535  *
 536  * NFSv4.0 callback argument types are defined in section 15 of RFC
 537  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
 538  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
 539  * Protocol".
 540  */
 541
 542 /*
 543  * NB: Without this zero space reservation, callbacks over krb5p fail
 544  */
 545 static void nfs4_xdr_enc_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
 546                                  const void *__unused)
 547 {
 548         xdr_reserve_space(xdr, 0);
 549 }
 550
 551 /*
 552  * 20.1.  Operation 3: CB_GETATTR - Get Attributes
 553  */
 554 static void nfs4_xdr_enc_cb_getattr(struct rpc_rqst *req,
 555                 struct xdr_stream *xdr, const void *data)
 556 {
 557         const struct nfsd4_callback *cb = data;
 558         struct nfs4_cb_fattr *ncf =
 559                 container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
 560         struct nfs4_cb_compound_hdr hdr = {
 561                 .ident = cb->cb_clp->cl_cb_ident,
 562                 .minorversion = cb->cb_clp->cl_minorversion,
 563         };
 564
 565         encode_cb_compound4args(xdr, &hdr);
 566         encode_cb_sequence4args(xdr, cb, &hdr);
 567         encode_cb_getattr4args(xdr, &hdr, ncf);
 568         encode_cb_nops(&hdr);
 569 }
 570
 571 /*
 572  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
 573  */
 574 static void nfs4_xdr_enc_cb_recall(struct rpc_rqst *req, struct xdr_stream *xdr,
 575                                    const void *data)
 576 {
 577         const struct nfsd4_callback *cb = data;
 578         const struct nfs4_delegation *dp = cb_to_delegation(cb);
 579         struct nfs4_cb_compound_hdr hdr = {
 580                 .ident = cb->cb_clp->cl_cb_ident,
 581                 .minorversion = cb->cb_clp->cl_minorversion,
 582         };
 583
 584         encode_cb_compound4args(xdr, &hdr);
 585         encode_cb_sequence4args(xdr, cb, &hdr);
 586         encode_cb_recall4args(xdr, dp, &hdr);
 587         encode_cb_nops(&hdr);
 588 }
 589
 590 /*
 591  * 20.6. Operation 8: CB_RECALL_ANY - Keep Any N Recallable Objects
 592  */
 593 static void
 594 nfs4_xdr_enc_cb_recall_any(struct rpc_rqst *req,
 595                 struct xdr_stream *xdr, const void *data)
 596 {
 597         const struct nfsd4_callback *cb = data;
 598         struct nfsd4_cb_recall_any *ra;
 599         struct nfs4_cb_compound_hdr hdr = {
 600                 .ident = cb->cb_clp->cl_cb_ident,
 601                 .minorversion = cb->cb_clp->cl_minorversion,
 602         };
 603
 604         ra = container_of(cb, struct nfsd4_cb_recall_any, ra_cb);
 605         encode_cb_compound4args(xdr, &hdr);
 606         encode_cb_sequence4args(xdr, cb, &hdr);
 607         encode_cb_recallany4args(xdr, &hdr, ra);
 608         encode_cb_nops(&hdr);
 609 }
 610
 611 /*
 612  * NFSv4.0 and NFSv4.1 XDR decode functions
 613  *
 614  * NFSv4.0 callback result types are defined in section 15 of RFC
 615  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
 616  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
 617  * Protocol".
 618  */
 619
 620 static int nfs4_xdr_dec_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
 621                                 void *__unused)
 622 {
 623         return 0;
 624 }
 625
 626 /*
 627  * 20.1.  Operation 3: CB_GETATTR - Get Attributes
 628  */
 629 static int nfs4_xdr_dec_cb_getattr(struct rpc_rqst *rqstp,
 630                                   struct xdr_stream *xdr,
 631                                   void *data)
 632 {
 633         struct nfsd4_callback *cb = data;
 634         struct nfs4_cb_compound_hdr hdr;
 635         int status;
 636         u32 bitmap[3] = {0};
 637         u32 attrlen;
 638         struct nfs4_cb_fattr *ncf =
 639                 container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
 640
 641         status = decode_cb_compound4res(xdr, &hdr);
 642         if (unlikely(status))
 643                 return status;
 644
 645         status = decode_cb_sequence4res(xdr, cb);
 646         if (unlikely(status || cb->cb_seq_status))
 647                 return status;
 648
 649         status = decode_cb_op_status(xdr, OP_CB_GETATTR, &cb->cb_status);
 650         if (status)
 651                 return status;
 652         if (xdr_stream_decode_uint32_array(xdr, bitmap, 3) < 0)
 653                 return -NFSERR_BAD_XDR;
 654         if (xdr_stream_decode_u32(xdr, &attrlen) < 0)
 655                 return -NFSERR_BAD_XDR;
 656         if (attrlen > (sizeof(ncf->ncf_cb_change) + sizeof(ncf->ncf_cb_fsize)))
 657                 return -NFSERR_BAD_XDR;
 658         status = decode_cb_fattr4(xdr, bitmap, ncf);
 659         return status;
 660 }
 661
 662 /*
 663  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
 664  */
 665 static int nfs4_xdr_dec_cb_recall(struct rpc_rqst *rqstp,
 666                                   struct xdr_stream *xdr,
 667                                   void *data)
 668 {
 669         struct nfsd4_callback *cb = data;
 670         struct nfs4_cb_compound_hdr hdr;
 671         int status;
 672
 673         status = decode_cb_compound4res(xdr, &hdr);
 674         if (unlikely(status))
 675                 return status;
 676
 677         status = decode_cb_sequence4res(xdr, cb);
 678         if (unlikely(status || cb->cb_seq_status))
 679                 return status;
 680
 681         return decode_cb_op_status(xdr, OP_CB_RECALL, &cb->cb_status);
 682 }
 683
 684 /*
 685  * 20.6. Operation 8: CB_RECALL_ANY - Keep Any N Recallable Objects
 686  */
 687 static int
 688 nfs4_xdr_dec_cb_recall_any(struct rpc_rqst *rqstp,
 689                                   struct xdr_stream *xdr,
 690                                   void *data)
 691 {
 692         struct nfsd4_callback *cb = data;
 693         struct nfs4_cb_compound_hdr hdr;
 694         int status;
 695
 696         status = decode_cb_compound4res(xdr, &hdr);
 697         if (unlikely(status))
 698                 return status;
 699         status = decode_cb_sequence4res(xdr, cb);
 700         if (unlikely(status || cb->cb_seq_status))
 701                 return status;
 702         status =  decode_cb_op_status(xdr, OP_CB_RECALL_ANY, &cb->cb_status);
 703         return status;
 704 }
 705
 706 #ifdef CONFIG_NFSD_PNFS
 707 /*
 708  * CB_LAYOUTRECALL4args
 709  *
 710  *      struct layoutrecall_file4 {
 711  *              nfs_fh4         lor_fh;
 712  *              offset4         lor_offset;
 713  *              length4         lor_length;
 714  *              stateid4        lor_stateid;
 715  *      };
 716  *
 717  *      union layoutrecall4 switch(layoutrecall_type4 lor_recalltype) {
 718  *      case LAYOUTRECALL4_FILE:
 719  *              layoutrecall_file4 lor_layout;
 720  *      case LAYOUTRECALL4_FSID:
 721  *              fsid4              lor_fsid;
 722  *      case LAYOUTRECALL4_ALL:
 723  *              void;
 724  *      };
 725  *
 726  *      struct CB_LAYOUTRECALL4args {
 727  *              layouttype4             clora_type;
 728  *              layoutiomode4           clora_iomode;
 729  *              bool                    clora_changed;
 730  *              layoutrecall4           clora_recall;
 731  *      };
 732  */
 733 static void encode_cb_layout4args(struct xdr_stream *xdr,
 734                                   const struct nfs4_layout_stateid *ls,
 735                                   struct nfs4_cb_compound_hdr *hdr)
 736 {
 737         __be32 *p;
 738
 739         BUG_ON(hdr->minorversion == 0);
 740
 741         p = xdr_reserve_space(xdr, 5 * 4);
 742         *p++ = cpu_to_be32(OP_CB_LAYOUTRECALL);
 743         *p++ = cpu_to_be32(ls->ls_layout_type);
 744         *p++ = cpu_to_be32(IOMODE_ANY);
 745         *p++ = cpu_to_be32(1);
 746         *p = cpu_to_be32(RETURN_FILE);
 747
 748         encode_nfs_fh4(xdr, &ls->ls_stid.sc_file->fi_fhandle);
 749
 750         p = xdr_reserve_space(xdr, 2 * 8);
 751         p = xdr_encode_hyper(p, 0);
 752         xdr_encode_hyper(p, NFS4_MAX_UINT64);
 753
 754         encode_stateid4(xdr, &ls->ls_recall_sid);
 755
 756         hdr->nops++;
 757 }
 758
 759 static void nfs4_xdr_enc_cb_layout(struct rpc_rqst *req,
 760                                    struct xdr_stream *xdr,
 761                                    const void *data)
 762 {
 763         const struct nfsd4_callback *cb = data;
 764         const struct nfs4_layout_stateid *ls =
 765                 container_of(cb, struct nfs4_layout_stateid, ls_recall);
 766         struct nfs4_cb_compound_hdr hdr = {
 767                 .ident = 0,
 768                 .minorversion = cb->cb_clp->cl_minorversion,
 769         };
 770
 771         encode_cb_compound4args(xdr, &hdr);
 772         encode_cb_sequence4args(xdr, cb, &hdr);
 773         encode_cb_layout4args(xdr, ls, &hdr);
 774         encode_cb_nops(&hdr);
 775 }
 776
 777 static int nfs4_xdr_dec_cb_layout(struct rpc_rqst *rqstp,
 778                                   struct xdr_stream *xdr,
 779                                   void *data)
 780 {
 781         struct nfsd4_callback *cb = data;
 782         struct nfs4_cb_compound_hdr hdr;
 783         int status;
 784
 785         status = decode_cb_compound4res(xdr, &hdr);
 786         if (unlikely(status))
 787                 return status;
 788
 789         status = decode_cb_sequence4res(xdr, cb);
 790         if (unlikely(status || cb->cb_seq_status))
 791                 return status;
 792
 793         return decode_cb_op_status(xdr, OP_CB_LAYOUTRECALL, &cb->cb_status);
 794 }
 795 #endif /* CONFIG_NFSD_PNFS */
 796
 797 static void encode_stateowner(struct xdr_stream *xdr, struct nfs4_stateowner *so)
 798 {
 799         __be32  *p;
 800
 801         p = xdr_reserve_space(xdr, 8 + 4 + so->so_owner.len);
 802         p = xdr_encode_opaque_fixed(p, &so->so_client->cl_clientid, 8);
 803         xdr_encode_opaque(p, so->so_owner.data, so->so_owner.len);
 804 }
 805
 806 static void nfs4_xdr_enc_cb_notify_lock(struct rpc_rqst *req,
 807                                         struct xdr_stream *xdr,
 808                                         const void *data)
 809 {
 810         const struct nfsd4_callback *cb = data;
 811         const struct nfsd4_blocked_lock *nbl =
 812                 container_of(cb, struct nfsd4_blocked_lock, nbl_cb);
 813         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)nbl->nbl_lock.c.flc_owner;
 814         struct nfs4_cb_compound_hdr hdr = {
 815                 .ident = 0,
 816                 .minorversion = cb->cb_clp->cl_minorversion,
 817         };
 818
 819         __be32 *p;
 820
 821         BUG_ON(hdr.minorversion == 0);
 822
 823         encode_cb_compound4args(xdr, &hdr);
 824         encode_cb_sequence4args(xdr, cb, &hdr);
 825
 826         p = xdr_reserve_space(xdr, 4);
 827         *p = cpu_to_be32(OP_CB_NOTIFY_LOCK);
 828         encode_nfs_fh4(xdr, &nbl->nbl_fh);
 829         encode_stateowner(xdr, &lo->lo_owner);
 830         hdr.nops++;
 831
 832         encode_cb_nops(&hdr);
 833 }
 834
 835 static int nfs4_xdr_dec_cb_notify_lock(struct rpc_rqst *rqstp,
 836                                         struct xdr_stream *xdr,
 837                                         void *data)
 838 {
 839         struct nfsd4_callback *cb = data;
 840         struct nfs4_cb_compound_hdr hdr;
 841         int status;
 842
 843         status = decode_cb_compound4res(xdr, &hdr);
 844         if (unlikely(status))
 845                 return status;
 846
 847         status = decode_cb_sequence4res(xdr, cb);
 848         if (unlikely(status || cb->cb_seq_status))
 849                 return status;
 850
 851         return decode_cb_op_status(xdr, OP_CB_NOTIFY_LOCK, &cb->cb_status);
 852 }
 853
 854 /*
 855  * struct write_response4 {
 856  *      stateid4        wr_callback_id<1>;
 857  *      length4         wr_count;
 858  *      stable_how4     wr_committed;
 859  *      verifier4       wr_writeverf;
 860  * };
 861  * union offload_info4 switch (nfsstat4 coa_status) {
 862  *      case NFS4_OK:
 863  *              write_response4 coa_resok4;
 864  *      default:
 865  *              length4         coa_bytes_copied;
 866  * };
 867  * struct CB_OFFLOAD4args {
 868  *      nfs_fh4         coa_fh;
 869  *      stateid4        coa_stateid;
 870  *      offload_info4   coa_offload_info;
 871  * };
 872  */
 873 static void encode_offload_info4(struct xdr_stream *xdr,
 874                                  const struct nfsd4_cb_offload *cbo)
 875 {
 876         __be32 *p;
 877
 878         p = xdr_reserve_space(xdr, 4);
 879         *p = cbo->co_nfserr;
 880         switch (cbo->co_nfserr) {
 881         case nfs_ok:
 882                 p = xdr_reserve_space(xdr, 4 + 8 + 4 + NFS4_VERIFIER_SIZE);
 883                 p = xdr_encode_empty_array(p);
 884                 p = xdr_encode_hyper(p, cbo->co_res.wr_bytes_written);
 885                 *p++ = cpu_to_be32(cbo->co_res.wr_stable_how);
 886                 p = xdr_encode_opaque_fixed(p, cbo->co_res.wr_verifier.data,
 887                                             NFS4_VERIFIER_SIZE);
 888                 break;
 889         default:
 890                 p = xdr_reserve_space(xdr, 8);
 891                 /* We always return success if bytes were written */
 892                 p = xdr_encode_hyper(p, 0);
 893         }
 894 }
 895
 896 static void encode_cb_offload4args(struct xdr_stream *xdr,
 897                                    const struct nfsd4_cb_offload *cbo,
 898                                    struct nfs4_cb_compound_hdr *hdr)
 899 {
 900         __be32 *p;
 901
 902         p = xdr_reserve_space(xdr, 4);
 903         *p = cpu_to_be32(OP_CB_OFFLOAD);
 904         encode_nfs_fh4(xdr, &cbo->co_fh);
 905         encode_stateid4(xdr, &cbo->co_res.cb_stateid);
 906         encode_offload_info4(xdr, cbo);
 907
 908         hdr->nops++;
 909 }
 910
 911 static void nfs4_xdr_enc_cb_offload(struct rpc_rqst *req,
 912                                     struct xdr_stream *xdr,
 913                                     const void *data)
 914 {
 915         const struct nfsd4_callback *cb = data;
 916         const struct nfsd4_cb_offload *cbo =
 917                 container_of(cb, struct nfsd4_cb_offload, co_cb);
 918         struct nfs4_cb_compound_hdr hdr = {
 919                 .ident = 0,
 920                 .minorversion = cb->cb_clp->cl_minorversion,
 921         };
 922
 923         encode_cb_compound4args(xdr, &hdr);
 924         encode_cb_sequence4args(xdr, cb, &hdr);
 925         encode_cb_offload4args(xdr, cbo, &hdr);
 926         encode_cb_nops(&hdr);
 927 }
 928
 929 static int nfs4_xdr_dec_cb_offload(struct rpc_rqst *rqstp,
 930                                    struct xdr_stream *xdr,
 931                                    void *data)
 932 {
 933         struct nfsd4_callback *cb = data;
 934         struct nfs4_cb_compound_hdr hdr;
 935         int status;
 936
 937         status = decode_cb_compound4res(xdr, &hdr);
 938         if (unlikely(status))
 939                 return status;
 940
 941         status = decode_cb_sequence4res(xdr, cb);
 942         if (unlikely(status || cb->cb_seq_status))
 943                 return status;
 944
 945         return decode_cb_op_status(xdr, OP_CB_OFFLOAD, &cb->cb_status);
 946 }
 947 /*
 948  * RPC procedure tables
 949  */
 950 #define PROC(proc, call, argtype, restype)                              \
 951 [NFSPROC4_CLNT_##proc] = {                                              \
 952         .p_proc    = NFSPROC4_CB_##call,                                \
 953         .p_encode  = nfs4_xdr_enc_##argtype,            \
 954         .p_decode  = nfs4_xdr_dec_##restype,                            \
 955         .p_arglen  = NFS4_enc_##argtype##_sz,                           \
 956         .p_replen  = NFS4_dec_##restype##_sz,                           \
 957         .p_statidx = NFSPROC4_CB_##call,                                \
 958         .p_name    = #proc,                                             \
 959 }
 960
 961 static const struct rpc_procinfo nfs4_cb_procedures[] = {
 962         PROC(CB_NULL,   NULL,           cb_null,        cb_null),
 963         PROC(CB_RECALL, COMPOUND,       cb_recall,      cb_recall),
 964 #ifdef CONFIG_NFSD_PNFS
 965         PROC(CB_LAYOUT, COMPOUND,       cb_layout,      cb_layout),
 966 #endif
 967         PROC(CB_NOTIFY_LOCK,    COMPOUND,       cb_notify_lock, cb_notify_lock),
 968         PROC(CB_OFFLOAD,        COMPOUND,       cb_offload,     cb_offload),
 969         PROC(CB_RECALL_ANY,     COMPOUND,       cb_recall_any,  cb_recall_any),
 970         PROC(CB_GETATTR,        COMPOUND,       cb_getattr,     cb_getattr),
 971 };
 972
 973 static unsigned int nfs4_cb_counts[ARRAY_SIZE(nfs4_cb_procedures)];
 974 static const struct rpc_version nfs_cb_version4 = {
 975 /*
 976  * Note on the callback rpc program version number: despite language in rfc
 977  * 5661 section 18.36.3 requiring servers to use 4 in this field, the
 978  * official xdr descriptions for both 4.0 and 4.1 specify version 1, and
 979  * in practice that appears to be what implementations use.  The section
 980  * 18.36.3 language is expected to be fixed in an erratum.
 981  */
 982         .number                 = 1,
 983         .nrprocs                = ARRAY_SIZE(nfs4_cb_procedures),
 984         .procs                  = nfs4_cb_procedures,
 985         .counts                 = nfs4_cb_counts,
 986 };
 987
 988 static const struct rpc_version *nfs_cb_version[2] = {
 989         [1] = &nfs_cb_version4,
 990 };
 991
 992 static const struct rpc_program cb_program;
 993
 994 static struct rpc_stat cb_stats = {
 995         .program                = &cb_program
 996 };
 997
 998 #define NFS4_CALLBACK 0x40000000
 999 static const struct rpc_program cb_program = {
1000         .name                   = "nfs4_cb",
1001         .number                 = NFS4_CALLBACK,
1002         .nrvers                 = ARRAY_SIZE(nfs_cb_version),
1003         .version                = nfs_cb_version,
1004         .stats                  = &cb_stats,
1005         .pipe_dir_name          = "nfsd4_cb",
1006 };
1007
1008 static int max_cb_time(struct net *net)
1009 {
1010         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1011
1012         /*
1013          * nfsd4_lease is set to at most one hour in __nfsd4_write_time,
1014          * so we can use 32-bit math on it. Warn if that assumption
1015          * ever stops being true.
1016          */
1017         if (WARN_ON_ONCE(nn->nfsd4_lease > 3600))
1018                 return 360 * HZ;
1019
1020         return max(((u32)nn->nfsd4_lease)/10, 1u) * HZ;
1021 }
1022
1023 static bool nfsd4_queue_cb(struct nfsd4_callback *cb)
1024 {
1025         struct nfs4_client *clp = cb->cb_clp;
1026
1027         trace_nfsd_cb_queue(clp, cb);
1028         return queue_work(clp->cl_callback_wq, &cb->cb_work);
1029 }
1030
1031 static void nfsd41_cb_inflight_begin(struct nfs4_client *clp)
1032 {
1033         atomic_inc(&clp->cl_cb_inflight);
1034 }
1035
1036 static void nfsd41_cb_inflight_end(struct nfs4_client *clp)
1037 {
1038
1039         if (atomic_dec_and_test(&clp->cl_cb_inflight))
1040                 wake_up_var(&clp->cl_cb_inflight);
1041 }
1042
1043 static void nfsd41_cb_inflight_wait_complete(struct nfs4_client *clp)
1044 {
1045         wait_var_event(&clp->cl_cb_inflight,
1046                         !atomic_read(&clp->cl_cb_inflight));
1047 }
1048
1049 static const struct cred *get_backchannel_cred(struct nfs4_client *clp, struct rpc_clnt *client, struct nfsd4_session *ses)
1050 {
1051         if (clp->cl_minorversion == 0) {
1052                 client->cl_principal = clp->cl_cred.cr_targ_princ ?
1053                         clp->cl_cred.cr_targ_princ : "nfs";
1054
1055                 return get_cred(rpc_machine_cred());
1056         } else {
1057                 struct cred *kcred;
1058
1059                 kcred = prepare_kernel_cred(&init_task);
1060                 if (!kcred)
1061                         return NULL;
1062
1063                 kcred->fsuid = ses->se_cb_sec.uid;
1064                 kcred->fsgid = ses->se_cb_sec.gid;
1065                 return kcred;
1066         }
1067 }
1068
1069 static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn, struct nfsd4_session *ses)
1070 {
1071         int maxtime = max_cb_time(clp->net);
1072         struct rpc_timeout      timeparms = {
1073                 .to_initval     = maxtime,
1074                 .to_retries     = 0,
1075                 .to_maxval      = maxtime,
1076         };
1077         struct rpc_create_args args = {
1078                 .net            = clp->net,
1079                 .address        = (struct sockaddr *) &conn->cb_addr,
1080                 .addrsize       = conn->cb_addrlen,
1081                 .saddress       = (struct sockaddr *) &conn->cb_saddr,
1082                 .timeout        = &timeparms,
1083                 .program        = &cb_program,
1084                 .version        = 1,
1085                 .flags          = (RPC_CLNT_CREATE_NOPING | RPC_CLNT_CREATE_QUIET),
1086                 .cred           = current_cred(),
1087         };
1088         struct rpc_clnt *client;
1089         const struct cred *cred;
1090
1091         if (clp->cl_minorversion == 0) {
1092                 if (!clp->cl_cred.cr_principal &&
1093                     (clp->cl_cred.cr_flavor >= RPC_AUTH_GSS_KRB5)) {
1094                         trace_nfsd_cb_setup_err(clp, -EINVAL);
1095                         return -EINVAL;
1096                 }
1097                 args.client_name = clp->cl_cred.cr_principal;
1098                 args.prognumber = conn->cb_prog;
1099                 args.protocol = XPRT_TRANSPORT_TCP;
1100                 args.authflavor = clp->cl_cred.cr_flavor;
1101                 clp->cl_cb_ident = conn->cb_ident;
1102         } else {
1103                 if (!conn->cb_xprt || !ses)
1104                         return -EINVAL;
1105                 clp->cl_cb_session = ses;
1106                 args.bc_xprt = conn->cb_xprt;
1107                 args.prognumber = clp->cl_cb_session->se_cb_prog;
1108                 args.protocol = conn->cb_xprt->xpt_class->xcl_ident |
1109                                 XPRT_TRANSPORT_BC;
1110                 args.authflavor = ses->se_cb_sec.flavor;
1111         }
1112         /* Create RPC client */
1113         client = rpc_create(&args);
1114         if (IS_ERR(client)) {
1115                 trace_nfsd_cb_setup_err(clp, PTR_ERR(client));
1116                 return PTR_ERR(client);
1117         }
1118         cred = get_backchannel_cred(clp, client, ses);
1119         if (!cred) {
1120                 trace_nfsd_cb_setup_err(clp, -ENOMEM);
1121                 rpc_shutdown_client(client);
1122                 return -ENOMEM;
1123         }
1124
1125         if (clp->cl_minorversion != 0)
1126                 clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
1127         clp->cl_cb_client = client;
1128         clp->cl_cb_cred = cred;
1129         rcu_read_lock();
1130         trace_nfsd_cb_setup(clp, rpc_peeraddr2str(client, RPC_DISPLAY_NETID),
1131                             args.authflavor);
1132         rcu_read_unlock();
1133         return 0;
1134 }
1135
1136 static void nfsd4_mark_cb_state(struct nfs4_client *clp, int newstate)
1137 {
1138         if (clp->cl_cb_state != newstate) {
1139                 clp->cl_cb_state = newstate;
1140                 trace_nfsd_cb_new_state(clp);
1141         }
1142 }
1143
1144 static void nfsd4_mark_cb_down(struct nfs4_client *clp)
1145 {
1146         if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
1147                 return;
1148         nfsd4_mark_cb_state(clp, NFSD4_CB_DOWN);
1149 }
1150
1151 static void nfsd4_mark_cb_fault(struct nfs4_client *clp)
1152 {
1153         if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
1154                 return;
1155         nfsd4_mark_cb_state(clp, NFSD4_CB_FAULT);
1156 }
1157
1158 static void nfsd4_cb_probe_done(struct rpc_task *task, void *calldata)
1159 {
1160         struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
1161
1162         if (task->tk_status)
1163                 nfsd4_mark_cb_down(clp);
1164         else
1165                 nfsd4_mark_cb_state(clp, NFSD4_CB_UP);
1166 }
1167
1168 static void nfsd4_cb_probe_release(void *calldata)
1169 {
1170         struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
1171
1172         nfsd41_cb_inflight_end(clp);
1173
1174 }
1175
1176 static const struct rpc_call_ops nfsd4_cb_probe_ops = {
1177         /* XXX: release method to ensure we set the cb channel down if
1178          * necessary on early failure? */
1179         .rpc_call_done = nfsd4_cb_probe_done,
1180         .rpc_release = nfsd4_cb_probe_release,
1181 };
1182
1183 /*
1184  * Poke the callback thread to process any updates to the callback
1185  * parameters, and send a null probe.
1186  */
1187 void nfsd4_probe_callback(struct nfs4_client *clp)
1188 {
1189         trace_nfsd_cb_probe(clp);
1190         nfsd4_mark_cb_state(clp, NFSD4_CB_UNKNOWN);
1191         set_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
1192         nfsd4_run_cb(&clp->cl_cb_null);
1193 }
1194
1195 void nfsd4_probe_callback_sync(struct nfs4_client *clp)
1196 {
1197         nfsd4_probe_callback(clp);
1198         flush_workqueue(clp->cl_callback_wq);
1199 }
1200
1201 void nfsd4_change_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
1202 {
1203         nfsd4_mark_cb_state(clp, NFSD4_CB_UNKNOWN);
1204         spin_lock(&clp->cl_lock);
1205         memcpy(&clp->cl_cb_conn, conn, sizeof(struct nfs4_cb_conn));
1206         spin_unlock(&clp->cl_lock);
1207 }
1208
1209 static int grab_slot(struct nfsd4_session *ses)
1210 {
1211         int idx;
1212
1213         spin_lock(&ses->se_lock);
1214         idx = ffs(ses->se_cb_slot_avail) - 1;
1215         if (idx < 0 || idx > ses->se_cb_highest_slot) {
1216                 spin_unlock(&ses->se_lock);
1217                 return -1;
1218         }
1219         /* clear the bit for the slot */
1220         ses->se_cb_slot_avail &= ~BIT(idx);
1221         spin_unlock(&ses->se_lock);
1222         return idx;
1223 }
1224
1225 /*
1226  * There's currently a single callback channel slot.
1227  * If the slot is available, then mark it busy.  Otherwise, set the
1228  * thread for sleeping on the callback RPC wait queue.
1229  */
1230 static bool nfsd41_cb_get_slot(struct nfsd4_callback *cb, struct rpc_task *task)
1231 {
1232         struct nfs4_client *clp = cb->cb_clp;
1233         struct nfsd4_session *ses = clp->cl_cb_session;
1234
1235         if (cb->cb_held_slot >= 0)
1236                 return true;
1237         cb->cb_held_slot = grab_slot(ses);
1238         if (cb->cb_held_slot < 0) {
1239                 rpc_sleep_on(&clp->cl_cb_waitq, task, NULL);
1240                 /* Race breaker */
1241                 cb->cb_held_slot = grab_slot(ses);
1242                 if (cb->cb_held_slot < 0)
1243                         return false;
1244                 rpc_wake_up_queued_task(&clp->cl_cb_waitq, task);
1245         }
1246         return true;
1247 }
1248
1249 static void nfsd41_cb_release_slot(struct nfsd4_callback *cb)
1250 {
1251         struct nfs4_client *clp = cb->cb_clp;
1252         struct nfsd4_session *ses = clp->cl_cb_session;
1253
1254         if (cb->cb_held_slot >= 0) {
1255                 spin_lock(&ses->se_lock);
1256                 ses->se_cb_slot_avail |= BIT(cb->cb_held_slot);
1257                 spin_unlock(&ses->se_lock);
1258                 cb->cb_held_slot = -1;
1259                 rpc_wake_up_next(&clp->cl_cb_waitq);
1260         }
1261 }
1262
1263 static void nfsd41_destroy_cb(struct nfsd4_callback *cb)
1264 {
1265         struct nfs4_client *clp = cb->cb_clp;
1266
1267         trace_nfsd_cb_destroy(clp, cb);
1268         nfsd41_cb_release_slot(cb);
1269         if (cb->cb_ops && cb->cb_ops->release)
1270                 cb->cb_ops->release(cb);
1271         nfsd41_cb_inflight_end(clp);
1272 }
1273
1274 /*
1275  * TODO: cb_sequence should support referring call lists, cachethis,
1276  * and mark callback channel down on communication errors.
1277  */
1278 static void nfsd4_cb_prepare(struct rpc_task *task, void *calldata)
1279 {
1280         struct nfsd4_callback *cb = calldata;
1281         struct nfs4_client *clp = cb->cb_clp;
1282         u32 minorversion = clp->cl_minorversion;
1283
1284         /*
1285          * cb_seq_status is only set in decode_cb_sequence4res,
1286          * and so will remain 1 if an rpc level failure occurs.
1287          */
1288         trace_nfsd_cb_rpc_prepare(clp);
1289         cb->cb_seq_status = 1;
1290         cb->cb_status = 0;
1291         if (minorversion && !nfsd41_cb_get_slot(cb, task))
1292                 return;
1293         rpc_call_start(task);
1294 }
1295
1296 static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback *cb)
1297 {
1298         struct nfs4_client *clp = cb->cb_clp;
1299         struct nfsd4_session *session = clp->cl_cb_session;
1300         bool ret = true;
1301
1302         if (!clp->cl_minorversion) {
1303                 /*
1304                  * If the backchannel connection was shut down while this
1305                  * task was queued, we need to resubmit it after setting up
1306                  * a new backchannel connection.
1307                  *
1308                  * Note that if we lost our callback connection permanently
1309                  * the submission code will error out, so we don't need to
1310                  * handle that case here.
1311                  */
1312                 if (RPC_SIGNALLED(task))
1313                         goto need_restart;
1314
1315                 return true;
1316         }
1317
1318         if (cb->cb_held_slot < 0)
1319                 goto need_restart;
1320
1321         /* This is the operation status code for CB_SEQUENCE */
1322         trace_nfsd_cb_seq_status(task, cb);
1323         switch (cb->cb_seq_status) {
1324         case 0:
1325                 /*
1326                  * No need for lock, access serialized in nfsd4_cb_prepare
1327                  *
1328                  * RFC5661 20.9.3
1329                  * If CB_SEQUENCE returns an error, then the state of the slot
1330                  * (sequence ID, cached reply) MUST NOT change.
1331                  */
1332                 ++session->se_cb_seq_nr[cb->cb_held_slot];
1333                 break;
1334         case -ESERVERFAULT:
1335                 ++session->se_cb_seq_nr[cb->cb_held_slot];
1336                 nfsd4_mark_cb_fault(cb->cb_clp);
1337                 ret = false;
1338                 break;
1339         case 1:
1340                 /*
1341                  * cb_seq_status remains 1 if an RPC Reply was never
1342                  * received. NFSD can't know if the client processed
1343                  * the CB_SEQUENCE operation. Ask the client to send a
1344                  * DESTROY_SESSION to recover.
1345                  */
1346                 fallthrough;
1347         case -NFS4ERR_BADSESSION:
1348                 nfsd4_mark_cb_fault(cb->cb_clp);
1349                 ret = false;
1350                 goto need_restart;
1351         case -NFS4ERR_DELAY:
1352                 cb->cb_seq_status = 1;
1353                 if (!rpc_restart_call(task))
1354                         goto out;
1355
1356                 rpc_delay(task, 2 * HZ);
1357                 return false;
1358         case -NFS4ERR_BADSLOT:
1359                 goto retry_nowait;
1360         case -NFS4ERR_SEQ_MISORDERED:
1361                 if (session->se_cb_seq_nr[cb->cb_held_slot] != 1) {
1362                         session->se_cb_seq_nr[cb->cb_held_slot] = 1;
1363                         goto retry_nowait;
1364                 }
1365                 break;
1366         default:
1367                 nfsd4_mark_cb_fault(cb->cb_clp);
1368         }
1369         trace_nfsd_cb_free_slot(task, cb);
1370         nfsd41_cb_release_slot(cb);
1371
1372         if (RPC_SIGNALLED(task))
1373                 goto need_restart;
1374 out:
1375         return ret;
1376 retry_nowait:
1377         if (rpc_restart_call_prepare(task))
1378                 ret = false;
1379         goto out;
1380 need_restart:
1381         if (!test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags)) {
1382                 trace_nfsd_cb_restart(clp, cb);
1383                 task->tk_status = 0;
1384                 cb->cb_need_restart = true;
1385         }
1386         return false;
1387 }
1388
1389 static void nfsd4_cb_done(struct rpc_task *task, void *calldata)
1390 {
1391         struct nfsd4_callback *cb = calldata;
1392         struct nfs4_client *clp = cb->cb_clp;
1393
1394         trace_nfsd_cb_rpc_done(clp);
1395
1396         if (!nfsd4_cb_sequence_done(task, cb))
1397                 return;
1398
1399         if (cb->cb_status) {
1400                 WARN_ONCE(task->tk_status, "cb_status=%d tk_status=%d",
1401                           cb->cb_status, task->tk_status);
1402                 task->tk_status = cb->cb_status;
1403         }
1404
1405         switch (cb->cb_ops->done(cb, task)) {
1406         case 0:
1407                 task->tk_status = 0;
1408                 rpc_restart_call_prepare(task);
1409                 return;
1410         case 1:
1411                 switch (task->tk_status) {
1412                 case -EIO:
1413                 case -ETIMEDOUT:
1414                 case -EACCES:
1415                         nfsd4_mark_cb_down(clp);
1416                 }
1417                 break;
1418         default:
1419                 BUG();
1420         }
1421 }
1422
1423 static void nfsd4_cb_release(void *calldata)
1424 {
1425         struct nfsd4_callback *cb = calldata;
1426
1427         trace_nfsd_cb_rpc_release(cb->cb_clp);
1428
1429         if (cb->cb_need_restart)
1430                 nfsd4_queue_cb(cb);
1431         else
1432                 nfsd41_destroy_cb(cb);
1433
1434 }
1435
1436 static const struct rpc_call_ops nfsd4_cb_ops = {
1437         .rpc_call_prepare = nfsd4_cb_prepare,
1438         .rpc_call_done = nfsd4_cb_done,
1439         .rpc_release = nfsd4_cb_release,
1440 };
1441
1442 /* must be called under the state lock */
1443 void nfsd4_shutdown_callback(struct nfs4_client *clp)
1444 {
1445         if (clp->cl_cb_state != NFSD4_CB_UNKNOWN)
1446                 trace_nfsd_cb_shutdown(clp);
1447
1448         set_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags);
1449         /*
1450          * Note this won't actually result in a null callback;
1451          * instead, nfsd4_run_cb_null() will detect the killed
1452          * client, destroy the rpc client, and stop:
1453          */
1454         nfsd4_run_cb(&clp->cl_cb_null);
1455         flush_workqueue(clp->cl_callback_wq);
1456         nfsd41_cb_inflight_wait_complete(clp);
1457 }
1458
1459 static struct nfsd4_conn * __nfsd4_find_backchannel(struct nfs4_client *clp)
1460 {
1461         struct nfsd4_session *s;
1462         struct nfsd4_conn *c;
1463
1464         lockdep_assert_held(&clp->cl_lock);
1465
1466         list_for_each_entry(s, &clp->cl_sessions, se_perclnt) {
1467                 list_for_each_entry(c, &s->se_conns, cn_persession) {
1468                         if (c->cn_flags & NFS4_CDFC4_BACK)
1469                                 return c;
1470                 }
1471         }
1472         return NULL;
1473 }
1474
1475 /*
1476  * Note there isn't a lot of locking in this code; instead we depend on
1477  * the fact that it is run from clp->cl_callback_wq, which won't run two
1478  * work items at once.  So, for example, clp->cl_callback_wq handles all
1479  * access of cl_cb_client and all calls to rpc_create or rpc_shutdown_client.
1480  */
1481 static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
1482 {
1483         struct nfs4_cb_conn conn;
1484         struct nfs4_client *clp = cb->cb_clp;
1485         struct nfsd4_session *ses = NULL;
1486         struct nfsd4_conn *c;
1487         int err;
1488
1489         trace_nfsd_cb_bc_update(clp, cb);
1490
1491         /*
1492          * This is either an update, or the client dying; in either case,
1493          * kill the old client:
1494          */
1495         if (clp->cl_cb_client) {
1496                 trace_nfsd_cb_bc_shutdown(clp, cb);
1497                 rpc_shutdown_client(clp->cl_cb_client);
1498                 clp->cl_cb_client = NULL;
1499                 put_cred(clp->cl_cb_cred);
1500                 clp->cl_cb_cred = NULL;
1501         }
1502         if (clp->cl_cb_conn.cb_xprt) {
1503                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1504                 clp->cl_cb_conn.cb_xprt = NULL;
1505         }
1506         if (test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags))
1507                 return;
1508
1509         spin_lock(&clp->cl_lock);
1510         /*
1511          * Only serialized callback code is allowed to clear these
1512          * flags; main nfsd code can only set them:
1513          */
1514         WARN_ON(!(clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK));
1515         clear_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
1516
1517         memcpy(&conn, &cb->cb_clp->cl_cb_conn, sizeof(struct nfs4_cb_conn));
1518         c = __nfsd4_find_backchannel(clp);
1519         if (c) {
1520                 svc_xprt_get(c->cn_xprt);
1521                 conn.cb_xprt = c->cn_xprt;
1522                 ses = c->cn_session;
1523         }
1524         spin_unlock(&clp->cl_lock);
1525
1526         err = setup_callback_client(clp, &conn, ses);
1527         if (err) {
1528                 nfsd4_mark_cb_down(clp);
1529                 if (c)
1530                         svc_xprt_put(c->cn_xprt);
1531                 return;
1532         }
1533 }
1534
1535 static void
1536 nfsd4_run_cb_work(struct work_struct *work)
1537 {
1538         struct nfsd4_callback *cb =
1539                 container_of(work, struct nfsd4_callback, cb_work);
1540         struct nfs4_client *clp = cb->cb_clp;
1541         struct rpc_clnt *clnt;
1542         int flags;
1543
1544         trace_nfsd_cb_start(clp);
1545
1546         if (clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK)
1547                 nfsd4_process_cb_update(cb);
1548
1549         clnt = clp->cl_cb_client;
1550         if (!clnt) {
1551                 /* Callback channel broken, or client killed; give up: */
1552                 nfsd41_destroy_cb(cb);
1553                 return;
1554         }
1555
1556         /*
1557          * Don't send probe messages for 4.1 or later.
1558          */
1559         if (!cb->cb_ops && clp->cl_minorversion) {
1560                 nfsd4_mark_cb_state(clp, NFSD4_CB_UP);
1561                 nfsd41_destroy_cb(cb);
1562                 return;
1563         }
1564
1565         if (cb->cb_need_restart) {
1566                 cb->cb_need_restart = false;
1567         } else {
1568                 if (cb->cb_ops && cb->cb_ops->prepare)
1569                         cb->cb_ops->prepare(cb);
1570         }
1571         cb->cb_msg.rpc_cred = clp->cl_cb_cred;
1572         flags = clp->cl_minorversion ? RPC_TASK_NOCONNECT : RPC_TASK_SOFTCONN;
1573         rpc_call_async(clnt, &cb->cb_msg, RPC_TASK_SOFT | flags,
1574                         cb->cb_ops ? &nfsd4_cb_ops : &nfsd4_cb_probe_ops, cb);
1575 }
1576
1577 void nfsd4_init_cb(struct nfsd4_callback *cb, struct nfs4_client *clp,
1578                 const struct nfsd4_callback_ops *ops, enum nfsd4_cb_op op)
1579 {
1580         cb->cb_clp = clp;
1581         cb->cb_msg.rpc_proc = &nfs4_cb_procedures[op];
1582         cb->cb_msg.rpc_argp = cb;
1583         cb->cb_msg.rpc_resp = cb;
1584         cb->cb_ops = ops;
1585         INIT_WORK(&cb->cb_work, nfsd4_run_cb_work);
1586         cb->cb_status = 0;
1587         cb->cb_need_restart = false;
1588         cb->cb_held_slot = -1;
1589 }
1590
1591 /**
1592  * nfsd4_run_cb - queue up a callback job to run
1593  * @cb: callback to queue
1594  *
1595  * Kick off a callback to do its thing. Returns false if it was already
1596  * on a queue, true otherwise.
1597  */
1598 bool nfsd4_run_cb(struct nfsd4_callback *cb)
1599 {
1600         struct nfs4_client *clp = cb->cb_clp;
1601         bool queued;
1602
1603         nfsd41_cb_inflight_begin(clp);
1604         queued = nfsd4_queue_cb(cb);
1605         if (!queued)
1606                 nfsd41_cb_inflight_end(clp);
1607         return queued;
1608 }