Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / nfsd / nfs4callback.c
1 /*
2  *  Copyright (c) 2001 The Regents of the University of Michigan.
3  *  All rights reserved.
4  *
5  *  Kendrick Smith <kmsmith@umich.edu>
6  *  Andy Adamson <andros@umich.edu>
7  *
8  *  Redistribution and use in source and binary forms, with or without
9  *  modification, are permitted provided that the following conditions
10  *  are met:
11  *
12  *  1. Redistributions of source code must retain the above copyright
13  *     notice, this list of conditions and the following disclaimer.
14  *  2. Redistributions in binary form must reproduce the above copyright
15  *     notice, this list of conditions and the following disclaimer in the
16  *     documentation and/or other materials provided with the distribution.
17  *  3. Neither the name of the University nor the names of its
18  *     contributors may be used to endorse or promote products derived
19  *     from this software without specific prior written permission.
20  *
21  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33
34 #include <linux/sunrpc/clnt.h>
35 #include <linux/sunrpc/xprt.h>
36 #include <linux/sunrpc/svc_xprt.h>
37 #include <linux/slab.h>
38 #include "nfsd.h"
39 #include "state.h"
40 #include "netns.h"
41 #include "xdr4cb.h"
42
43 #define NFSDDBG_FACILITY                NFSDDBG_PROC
44
45 static void nfsd4_mark_cb_fault(struct nfs4_client *, int reason);
46
47 #define NFSPROC4_CB_NULL 0
48 #define NFSPROC4_CB_COMPOUND 1
49
50 /* Index of predefined Linux callback client operations */
51
52 struct nfs4_cb_compound_hdr {
53         /* args */
54         u32             ident;  /* minorversion 0 only */
55         u32             nops;
56         __be32          *nops_p;
57         u32             minorversion;
58         /* res */
59         int             status;
60 };
61
62 /*
63  * Handle decode buffer overflows out-of-line.
64  */
65 static void print_overflow_msg(const char *func, const struct xdr_stream *xdr)
66 {
67         dprintk("NFS: %s prematurely hit the end of our receive buffer. "
68                 "Remaining buffer length is %tu words.\n",
69                 func, xdr->end - xdr->p);
70 }
71
72 static __be32 *xdr_encode_empty_array(__be32 *p)
73 {
74         *p++ = xdr_zero;
75         return p;
76 }
77
78 /*
79  * Encode/decode NFSv4 CB basic data types
80  *
81  * Basic NFSv4 callback data types are defined in section 15 of RFC
82  * 3530: "Network File System (NFS) version 4 Protocol" and section
83  * 20 of RFC 5661: "Network File System (NFS) Version 4 Minor Version
84  * 1 Protocol"
85  */
86
87 /*
88  *      nfs_cb_opnum4
89  *
90  *      enum nfs_cb_opnum4 {
91  *              OP_CB_GETATTR           = 3,
92  *                ...
93  *      };
94  */
95 enum nfs_cb_opnum4 {
96         OP_CB_GETATTR                   = 3,
97         OP_CB_RECALL                    = 4,
98         OP_CB_LAYOUTRECALL              = 5,
99         OP_CB_NOTIFY                    = 6,
100         OP_CB_PUSH_DELEG                = 7,
101         OP_CB_RECALL_ANY                = 8,
102         OP_CB_RECALLABLE_OBJ_AVAIL      = 9,
103         OP_CB_RECALL_SLOT               = 10,
104         OP_CB_SEQUENCE                  = 11,
105         OP_CB_WANTS_CANCELLED           = 12,
106         OP_CB_NOTIFY_LOCK               = 13,
107         OP_CB_NOTIFY_DEVICEID           = 14,
108         OP_CB_ILLEGAL                   = 10044
109 };
110
111 static void encode_nfs_cb_opnum4(struct xdr_stream *xdr, enum nfs_cb_opnum4 op)
112 {
113         __be32 *p;
114
115         p = xdr_reserve_space(xdr, 4);
116         *p = cpu_to_be32(op);
117 }
118
119 /*
120  * nfs_fh4
121  *
122  *      typedef opaque nfs_fh4<NFS4_FHSIZE>;
123  */
124 static void encode_nfs_fh4(struct xdr_stream *xdr, const struct knfsd_fh *fh)
125 {
126         u32 length = fh->fh_size;
127         __be32 *p;
128
129         BUG_ON(length > NFS4_FHSIZE);
130         p = xdr_reserve_space(xdr, 4 + length);
131         xdr_encode_opaque(p, &fh->fh_base, length);
132 }
133
134 /*
135  * stateid4
136  *
137  *      struct stateid4 {
138  *              uint32_t        seqid;
139  *              opaque          other[12];
140  *      };
141  */
142 static void encode_stateid4(struct xdr_stream *xdr, const stateid_t *sid)
143 {
144         __be32 *p;
145
146         p = xdr_reserve_space(xdr, NFS4_STATEID_SIZE);
147         *p++ = cpu_to_be32(sid->si_generation);
148         xdr_encode_opaque_fixed(p, &sid->si_opaque, NFS4_STATEID_OTHER_SIZE);
149 }
150
151 /*
152  * sessionid4
153  *
154  *      typedef opaque sessionid4[NFS4_SESSIONID_SIZE];
155  */
156 static void encode_sessionid4(struct xdr_stream *xdr,
157                               const struct nfsd4_session *session)
158 {
159         __be32 *p;
160
161         p = xdr_reserve_space(xdr, NFS4_MAX_SESSIONID_LEN);
162         xdr_encode_opaque_fixed(p, session->se_sessionid.data,
163                                         NFS4_MAX_SESSIONID_LEN);
164 }
165
166 /*
167  * nfsstat4
168  */
169 static const struct {
170         int stat;
171         int errno;
172 } nfs_cb_errtbl[] = {
173         { NFS4_OK,              0               },
174         { NFS4ERR_PERM,         -EPERM          },
175         { NFS4ERR_NOENT,        -ENOENT         },
176         { NFS4ERR_IO,           -EIO            },
177         { NFS4ERR_NXIO,         -ENXIO          },
178         { NFS4ERR_ACCESS,       -EACCES         },
179         { NFS4ERR_EXIST,        -EEXIST         },
180         { NFS4ERR_XDEV,         -EXDEV          },
181         { NFS4ERR_NOTDIR,       -ENOTDIR        },
182         { NFS4ERR_ISDIR,        -EISDIR         },
183         { NFS4ERR_INVAL,        -EINVAL         },
184         { NFS4ERR_FBIG,         -EFBIG          },
185         { NFS4ERR_NOSPC,        -ENOSPC         },
186         { NFS4ERR_ROFS,         -EROFS          },
187         { NFS4ERR_MLINK,        -EMLINK         },
188         { NFS4ERR_NAMETOOLONG,  -ENAMETOOLONG   },
189         { NFS4ERR_NOTEMPTY,     -ENOTEMPTY      },
190         { NFS4ERR_DQUOT,        -EDQUOT         },
191         { NFS4ERR_STALE,        -ESTALE         },
192         { NFS4ERR_BADHANDLE,    -EBADHANDLE     },
193         { NFS4ERR_BAD_COOKIE,   -EBADCOOKIE     },
194         { NFS4ERR_NOTSUPP,      -ENOTSUPP       },
195         { NFS4ERR_TOOSMALL,     -ETOOSMALL      },
196         { NFS4ERR_SERVERFAULT,  -ESERVERFAULT   },
197         { NFS4ERR_BADTYPE,      -EBADTYPE       },
198         { NFS4ERR_LOCKED,       -EAGAIN         },
199         { NFS4ERR_RESOURCE,     -EREMOTEIO      },
200         { NFS4ERR_SYMLINK,      -ELOOP          },
201         { NFS4ERR_OP_ILLEGAL,   -EOPNOTSUPP     },
202         { NFS4ERR_DEADLOCK,     -EDEADLK        },
203         { -1,                   -EIO            }
204 };
205
206 /*
207  * If we cannot translate the error, the recovery routines should
208  * handle it.
209  *
210  * Note: remaining NFSv4 error codes have values > 10000, so should
211  * not conflict with native Linux error codes.
212  */
213 static int nfs_cb_stat_to_errno(int status)
214 {
215         int i;
216
217         for (i = 0; nfs_cb_errtbl[i].stat != -1; i++) {
218                 if (nfs_cb_errtbl[i].stat == status)
219                         return nfs_cb_errtbl[i].errno;
220         }
221
222         dprintk("NFSD: Unrecognized NFS CB status value: %u\n", status);
223         return -status;
224 }
225
226 static int decode_cb_op_status(struct xdr_stream *xdr, enum nfs_opnum4 expected,
227                                int *status)
228 {
229         __be32 *p;
230         u32 op;
231
232         p = xdr_inline_decode(xdr, 4 + 4);
233         if (unlikely(p == NULL))
234                 goto out_overflow;
235         op = be32_to_cpup(p++);
236         if (unlikely(op != expected))
237                 goto out_unexpected;
238         *status = nfs_cb_stat_to_errno(be32_to_cpup(p));
239         return 0;
240 out_overflow:
241         print_overflow_msg(__func__, xdr);
242         return -EIO;
243 out_unexpected:
244         dprintk("NFSD: Callback server returned operation %d but "
245                 "we issued a request for %d\n", op, expected);
246         return -EIO;
247 }
248
249 /*
250  * CB_COMPOUND4args
251  *
252  *      struct CB_COMPOUND4args {
253  *              utf8str_cs      tag;
254  *              uint32_t        minorversion;
255  *              uint32_t        callback_ident;
256  *              nfs_cb_argop4   argarray<>;
257  *      };
258 */
259 static void encode_cb_compound4args(struct xdr_stream *xdr,
260                                     struct nfs4_cb_compound_hdr *hdr)
261 {
262         __be32 * p;
263
264         p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4);
265         p = xdr_encode_empty_array(p);          /* empty tag */
266         *p++ = cpu_to_be32(hdr->minorversion);
267         *p++ = cpu_to_be32(hdr->ident);
268
269         hdr->nops_p = p;
270         *p = cpu_to_be32(hdr->nops);            /* argarray element count */
271 }
272
273 /*
274  * Update argarray element count
275  */
276 static void encode_cb_nops(struct nfs4_cb_compound_hdr *hdr)
277 {
278         BUG_ON(hdr->nops > NFS4_MAX_BACK_CHANNEL_OPS);
279         *hdr->nops_p = cpu_to_be32(hdr->nops);
280 }
281
282 /*
283  * CB_COMPOUND4res
284  *
285  *      struct CB_COMPOUND4res {
286  *              nfsstat4        status;
287  *              utf8str_cs      tag;
288  *              nfs_cb_resop4   resarray<>;
289  *      };
290  */
291 static int decode_cb_compound4res(struct xdr_stream *xdr,
292                                   struct nfs4_cb_compound_hdr *hdr)
293 {
294         u32 length;
295         __be32 *p;
296
297         p = xdr_inline_decode(xdr, 4 + 4);
298         if (unlikely(p == NULL))
299                 goto out_overflow;
300         hdr->status = be32_to_cpup(p++);
301         /* Ignore the tag */
302         length = be32_to_cpup(p++);
303         p = xdr_inline_decode(xdr, length + 4);
304         if (unlikely(p == NULL))
305                 goto out_overflow;
306         hdr->nops = be32_to_cpup(p);
307         return 0;
308 out_overflow:
309         print_overflow_msg(__func__, xdr);
310         return -EIO;
311 }
312
313 /*
314  * CB_RECALL4args
315  *
316  *      struct CB_RECALL4args {
317  *              stateid4        stateid;
318  *              bool            truncate;
319  *              nfs_fh4         fh;
320  *      };
321  */
322 static void encode_cb_recall4args(struct xdr_stream *xdr,
323                                   const struct nfs4_delegation *dp,
324                                   struct nfs4_cb_compound_hdr *hdr)
325 {
326         __be32 *p;
327
328         encode_nfs_cb_opnum4(xdr, OP_CB_RECALL);
329         encode_stateid4(xdr, &dp->dl_stid.sc_stateid);
330
331         p = xdr_reserve_space(xdr, 4);
332         *p++ = xdr_zero;                        /* truncate */
333
334         encode_nfs_fh4(xdr, &dp->dl_stid.sc_file->fi_fhandle);
335
336         hdr->nops++;
337 }
338
339 /*
340  * CB_SEQUENCE4args
341  *
342  *      struct CB_SEQUENCE4args {
343  *              sessionid4              csa_sessionid;
344  *              sequenceid4             csa_sequenceid;
345  *              slotid4                 csa_slotid;
346  *              slotid4                 csa_highest_slotid;
347  *              bool                    csa_cachethis;
348  *              referring_call_list4    csa_referring_call_lists<>;
349  *      };
350  */
351 static void encode_cb_sequence4args(struct xdr_stream *xdr,
352                                     const struct nfsd4_callback *cb,
353                                     struct nfs4_cb_compound_hdr *hdr)
354 {
355         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
356         __be32 *p;
357
358         if (hdr->minorversion == 0)
359                 return;
360
361         encode_nfs_cb_opnum4(xdr, OP_CB_SEQUENCE);
362         encode_sessionid4(xdr, session);
363
364         p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4 + 4);
365         *p++ = cpu_to_be32(session->se_cb_seq_nr);      /* csa_sequenceid */
366         *p++ = xdr_zero;                        /* csa_slotid */
367         *p++ = xdr_zero;                        /* csa_highest_slotid */
368         *p++ = xdr_zero;                        /* csa_cachethis */
369         xdr_encode_empty_array(p);              /* csa_referring_call_lists */
370
371         hdr->nops++;
372 }
373
374 /*
375  * CB_SEQUENCE4resok
376  *
377  *      struct CB_SEQUENCE4resok {
378  *              sessionid4      csr_sessionid;
379  *              sequenceid4     csr_sequenceid;
380  *              slotid4         csr_slotid;
381  *              slotid4         csr_highest_slotid;
382  *              slotid4         csr_target_highest_slotid;
383  *      };
384  *
385  *      union CB_SEQUENCE4res switch (nfsstat4 csr_status) {
386  *      case NFS4_OK:
387  *              CB_SEQUENCE4resok       csr_resok4;
388  *      default:
389  *              void;
390  *      };
391  *
392  * Our current back channel implmentation supports a single backchannel
393  * with a single slot.
394  */
395 static int decode_cb_sequence4resok(struct xdr_stream *xdr,
396                                     struct nfsd4_callback *cb)
397 {
398         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
399         struct nfs4_sessionid id;
400         int status;
401         __be32 *p;
402         u32 dummy;
403
404         status = -ESERVERFAULT;
405
406         /*
407          * If the server returns different values for sessionID, slotID or
408          * sequence number, the server is looney tunes.
409          */
410         p = xdr_inline_decode(xdr, NFS4_MAX_SESSIONID_LEN + 4 + 4 + 4 + 4);
411         if (unlikely(p == NULL))
412                 goto out_overflow;
413         memcpy(id.data, p, NFS4_MAX_SESSIONID_LEN);
414         if (memcmp(id.data, session->se_sessionid.data,
415                                         NFS4_MAX_SESSIONID_LEN) != 0) {
416                 dprintk("NFS: %s Invalid session id\n", __func__);
417                 goto out;
418         }
419         p += XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN);
420
421         dummy = be32_to_cpup(p++);
422         if (dummy != session->se_cb_seq_nr) {
423                 dprintk("NFS: %s Invalid sequence number\n", __func__);
424                 goto out;
425         }
426
427         dummy = be32_to_cpup(p++);
428         if (dummy != 0) {
429                 dprintk("NFS: %s Invalid slotid\n", __func__);
430                 goto out;
431         }
432
433         /*
434          * FIXME: process highest slotid and target highest slotid
435          */
436         status = 0;
437 out:
438         if (status)
439                 nfsd4_mark_cb_fault(cb->cb_clp, status);
440         return status;
441 out_overflow:
442         print_overflow_msg(__func__, xdr);
443         return -EIO;
444 }
445
446 static int decode_cb_sequence4res(struct xdr_stream *xdr,
447                                   struct nfsd4_callback *cb)
448 {
449         int status;
450
451         if (cb->cb_minorversion == 0)
452                 return 0;
453
454         status = decode_cb_op_status(xdr, OP_CB_SEQUENCE, &cb->cb_status);
455         if (unlikely(status || cb->cb_status))
456                 return status;
457
458         return decode_cb_sequence4resok(xdr, cb);
459 }
460
461 /*
462  * NFSv4.0 and NFSv4.1 XDR encode functions
463  *
464  * NFSv4.0 callback argument types are defined in section 15 of RFC
465  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
466  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
467  * Protocol".
468  */
469
470 /*
471  * NB: Without this zero space reservation, callbacks over krb5p fail
472  */
473 static void nfs4_xdr_enc_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
474                                  void *__unused)
475 {
476         xdr_reserve_space(xdr, 0);
477 }
478
479 /*
480  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
481  */
482 static void nfs4_xdr_enc_cb_recall(struct rpc_rqst *req, struct xdr_stream *xdr,
483                                    const struct nfsd4_callback *cb)
484 {
485         const struct nfs4_delegation *dp = cb_to_delegation(cb);
486         struct nfs4_cb_compound_hdr hdr = {
487                 .ident = cb->cb_clp->cl_cb_ident,
488                 .minorversion = cb->cb_minorversion,
489         };
490
491         encode_cb_compound4args(xdr, &hdr);
492         encode_cb_sequence4args(xdr, cb, &hdr);
493         encode_cb_recall4args(xdr, dp, &hdr);
494         encode_cb_nops(&hdr);
495 }
496
497
498 /*
499  * NFSv4.0 and NFSv4.1 XDR decode functions
500  *
501  * NFSv4.0 callback result types are defined in section 15 of RFC
502  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
503  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
504  * Protocol".
505  */
506
507 static int nfs4_xdr_dec_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
508                                 void *__unused)
509 {
510         return 0;
511 }
512
513 /*
514  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
515  */
516 static int nfs4_xdr_dec_cb_recall(struct rpc_rqst *rqstp,
517                                   struct xdr_stream *xdr,
518                                   struct nfsd4_callback *cb)
519 {
520         struct nfs4_cb_compound_hdr hdr;
521         int status;
522
523         status = decode_cb_compound4res(xdr, &hdr);
524         if (unlikely(status))
525                 return status;
526
527         if (cb != NULL) {
528                 status = decode_cb_sequence4res(xdr, cb);
529                 if (unlikely(status || cb->cb_status))
530                         return status;
531         }
532
533         return decode_cb_op_status(xdr, OP_CB_RECALL, &cb->cb_status);
534 }
535
536 #ifdef CONFIG_NFSD_PNFS
537 /*
538  * CB_LAYOUTRECALL4args
539  *
540  *      struct layoutrecall_file4 {
541  *              nfs_fh4         lor_fh;
542  *              offset4         lor_offset;
543  *              length4         lor_length;
544  *              stateid4        lor_stateid;
545  *      };
546  *
547  *      union layoutrecall4 switch(layoutrecall_type4 lor_recalltype) {
548  *      case LAYOUTRECALL4_FILE:
549  *              layoutrecall_file4 lor_layout;
550  *      case LAYOUTRECALL4_FSID:
551  *              fsid4              lor_fsid;
552  *      case LAYOUTRECALL4_ALL:
553  *              void;
554  *      };
555  *
556  *      struct CB_LAYOUTRECALL4args {
557  *              layouttype4             clora_type;
558  *              layoutiomode4           clora_iomode;
559  *              bool                    clora_changed;
560  *              layoutrecall4           clora_recall;
561  *      };
562  */
563 static void encode_cb_layout4args(struct xdr_stream *xdr,
564                                   const struct nfs4_layout_stateid *ls,
565                                   struct nfs4_cb_compound_hdr *hdr)
566 {
567         __be32 *p;
568
569         BUG_ON(hdr->minorversion == 0);
570
571         p = xdr_reserve_space(xdr, 5 * 4);
572         *p++ = cpu_to_be32(OP_CB_LAYOUTRECALL);
573         *p++ = cpu_to_be32(ls->ls_layout_type);
574         *p++ = cpu_to_be32(IOMODE_ANY);
575         *p++ = cpu_to_be32(1);
576         *p = cpu_to_be32(RETURN_FILE);
577
578         encode_nfs_fh4(xdr, &ls->ls_stid.sc_file->fi_fhandle);
579
580         p = xdr_reserve_space(xdr, 2 * 8);
581         p = xdr_encode_hyper(p, 0);
582         xdr_encode_hyper(p, NFS4_MAX_UINT64);
583
584         encode_stateid4(xdr, &ls->ls_recall_sid);
585
586         hdr->nops++;
587 }
588
589 static void nfs4_xdr_enc_cb_layout(struct rpc_rqst *req,
590                                    struct xdr_stream *xdr,
591                                    const struct nfsd4_callback *cb)
592 {
593         const struct nfs4_layout_stateid *ls =
594                 container_of(cb, struct nfs4_layout_stateid, ls_recall);
595         struct nfs4_cb_compound_hdr hdr = {
596                 .ident = 0,
597                 .minorversion = cb->cb_minorversion,
598         };
599
600         encode_cb_compound4args(xdr, &hdr);
601         encode_cb_sequence4args(xdr, cb, &hdr);
602         encode_cb_layout4args(xdr, ls, &hdr);
603         encode_cb_nops(&hdr);
604 }
605
606 static int nfs4_xdr_dec_cb_layout(struct rpc_rqst *rqstp,
607                                   struct xdr_stream *xdr,
608                                   struct nfsd4_callback *cb)
609 {
610         struct nfs4_cb_compound_hdr hdr;
611         int status;
612
613         status = decode_cb_compound4res(xdr, &hdr);
614         if (unlikely(status))
615                 return status;
616
617         if (cb) {
618                 status = decode_cb_sequence4res(xdr, cb);
619                 if (unlikely(status || cb->cb_status))
620                         return status;
621         }
622         return decode_cb_op_status(xdr, OP_CB_LAYOUTRECALL, &cb->cb_status);
623 }
624 #endif /* CONFIG_NFSD_PNFS */
625
626 /*
627  * RPC procedure tables
628  */
629 #define PROC(proc, call, argtype, restype)                              \
630 [NFSPROC4_CLNT_##proc] = {                                              \
631         .p_proc    = NFSPROC4_CB_##call,                                \
632         .p_encode  = (kxdreproc_t)nfs4_xdr_enc_##argtype,               \
633         .p_decode  = (kxdrdproc_t)nfs4_xdr_dec_##restype,               \
634         .p_arglen  = NFS4_enc_##argtype##_sz,                           \
635         .p_replen  = NFS4_dec_##restype##_sz,                           \
636         .p_statidx = NFSPROC4_CB_##call,                                \
637         .p_name    = #proc,                                             \
638 }
639
640 static struct rpc_procinfo nfs4_cb_procedures[] = {
641         PROC(CB_NULL,   NULL,           cb_null,        cb_null),
642         PROC(CB_RECALL, COMPOUND,       cb_recall,      cb_recall),
643 #ifdef CONFIG_NFSD_PNFS
644         PROC(CB_LAYOUT, COMPOUND,       cb_layout,      cb_layout),
645 #endif
646 };
647
648 static struct rpc_version nfs_cb_version4 = {
649 /*
650  * Note on the callback rpc program version number: despite language in rfc
651  * 5661 section 18.36.3 requiring servers to use 4 in this field, the
652  * official xdr descriptions for both 4.0 and 4.1 specify version 1, and
653  * in practice that appears to be what implementations use.  The section
654  * 18.36.3 language is expected to be fixed in an erratum.
655  */
656         .number                 = 1,
657         .nrprocs                = ARRAY_SIZE(nfs4_cb_procedures),
658         .procs                  = nfs4_cb_procedures
659 };
660
661 static const struct rpc_version *nfs_cb_version[] = {
662         &nfs_cb_version4,
663 };
664
665 static const struct rpc_program cb_program;
666
667 static struct rpc_stat cb_stats = {
668         .program                = &cb_program
669 };
670
671 #define NFS4_CALLBACK 0x40000000
672 static const struct rpc_program cb_program = {
673         .name                   = "nfs4_cb",
674         .number                 = NFS4_CALLBACK,
675         .nrvers                 = ARRAY_SIZE(nfs_cb_version),
676         .version                = nfs_cb_version,
677         .stats                  = &cb_stats,
678         .pipe_dir_name          = "nfsd4_cb",
679 };
680
681 static int max_cb_time(struct net *net)
682 {
683         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
684         return max(nn->nfsd4_lease/10, (time_t)1) * HZ;
685 }
686
687 static struct rpc_cred *callback_cred;
688
689 int set_callback_cred(void)
690 {
691         if (callback_cred)
692                 return 0;
693         callback_cred = rpc_lookup_machine_cred("nfs");
694         if (!callback_cred)
695                 return -ENOMEM;
696         return 0;
697 }
698
699 static struct rpc_cred *get_backchannel_cred(struct nfs4_client *clp, struct rpc_clnt *client, struct nfsd4_session *ses)
700 {
701         if (clp->cl_minorversion == 0) {
702                 return get_rpccred(callback_cred);
703         } else {
704                 struct rpc_auth *auth = client->cl_auth;
705                 struct auth_cred acred = {};
706
707                 acred.uid = ses->se_cb_sec.uid;
708                 acred.gid = ses->se_cb_sec.gid;
709                 return auth->au_ops->lookup_cred(client->cl_auth, &acred, 0);
710         }
711 }
712
713 static struct rpc_clnt *create_backchannel_client(struct rpc_create_args *args)
714 {
715         struct rpc_xprt *xprt;
716
717         if (args->protocol != XPRT_TRANSPORT_BC_TCP)
718                 return rpc_create(args);
719
720         xprt = args->bc_xprt->xpt_bc_xprt;
721         if (xprt) {
722                 xprt_get(xprt);
723                 return rpc_create_xprt(args, xprt);
724         }
725
726         return rpc_create(args);
727 }
728
729 static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn, struct nfsd4_session *ses)
730 {
731         int maxtime = max_cb_time(clp->net);
732         struct rpc_timeout      timeparms = {
733                 .to_initval     = maxtime,
734                 .to_retries     = 0,
735                 .to_maxval      = maxtime,
736         };
737         struct rpc_create_args args = {
738                 .net            = clp->net,
739                 .address        = (struct sockaddr *) &conn->cb_addr,
740                 .addrsize       = conn->cb_addrlen,
741                 .saddress       = (struct sockaddr *) &conn->cb_saddr,
742                 .timeout        = &timeparms,
743                 .program        = &cb_program,
744                 .version        = 0,
745                 .flags          = (RPC_CLNT_CREATE_NOPING | RPC_CLNT_CREATE_QUIET),
746         };
747         struct rpc_clnt *client;
748         struct rpc_cred *cred;
749
750         if (clp->cl_minorversion == 0) {
751                 if (!clp->cl_cred.cr_principal &&
752                                 (clp->cl_cred.cr_flavor >= RPC_AUTH_GSS_KRB5))
753                         return -EINVAL;
754                 args.client_name = clp->cl_cred.cr_principal;
755                 args.prognumber = conn->cb_prog;
756                 args.protocol = XPRT_TRANSPORT_TCP;
757                 args.authflavor = clp->cl_cred.cr_flavor;
758                 clp->cl_cb_ident = conn->cb_ident;
759         } else {
760                 if (!conn->cb_xprt)
761                         return -EINVAL;
762                 clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
763                 clp->cl_cb_session = ses;
764                 args.bc_xprt = conn->cb_xprt;
765                 args.prognumber = clp->cl_cb_session->se_cb_prog;
766                 args.protocol = conn->cb_xprt->xpt_class->xcl_ident |
767                                 XPRT_TRANSPORT_BC;
768                 args.authflavor = ses->se_cb_sec.flavor;
769         }
770         /* Create RPC client */
771         client = create_backchannel_client(&args);
772         if (IS_ERR(client)) {
773                 dprintk("NFSD: couldn't create callback client: %ld\n",
774                         PTR_ERR(client));
775                 return PTR_ERR(client);
776         }
777         cred = get_backchannel_cred(clp, client, ses);
778         if (IS_ERR(cred)) {
779                 rpc_shutdown_client(client);
780                 return PTR_ERR(cred);
781         }
782         clp->cl_cb_client = client;
783         clp->cl_cb_cred = cred;
784         return 0;
785 }
786
787 static void warn_no_callback_path(struct nfs4_client *clp, int reason)
788 {
789         dprintk("NFSD: warning: no callback path to client %.*s: error %d\n",
790                 (int)clp->cl_name.len, clp->cl_name.data, reason);
791 }
792
793 static void nfsd4_mark_cb_down(struct nfs4_client *clp, int reason)
794 {
795         clp->cl_cb_state = NFSD4_CB_DOWN;
796         warn_no_callback_path(clp, reason);
797 }
798
799 static void nfsd4_mark_cb_fault(struct nfs4_client *clp, int reason)
800 {
801         clp->cl_cb_state = NFSD4_CB_FAULT;
802         warn_no_callback_path(clp, reason);
803 }
804
805 static void nfsd4_cb_probe_done(struct rpc_task *task, void *calldata)
806 {
807         struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
808
809         if (task->tk_status)
810                 nfsd4_mark_cb_down(clp, task->tk_status);
811         else
812                 clp->cl_cb_state = NFSD4_CB_UP;
813 }
814
815 static const struct rpc_call_ops nfsd4_cb_probe_ops = {
816         /* XXX: release method to ensure we set the cb channel down if
817          * necessary on early failure? */
818         .rpc_call_done = nfsd4_cb_probe_done,
819 };
820
821 static struct workqueue_struct *callback_wq;
822
823 /*
824  * Poke the callback thread to process any updates to the callback
825  * parameters, and send a null probe.
826  */
827 void nfsd4_probe_callback(struct nfs4_client *clp)
828 {
829         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
830         set_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
831         nfsd4_run_cb(&clp->cl_cb_null);
832 }
833
834 void nfsd4_probe_callback_sync(struct nfs4_client *clp)
835 {
836         nfsd4_probe_callback(clp);
837         flush_workqueue(callback_wq);
838 }
839
840 void nfsd4_change_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
841 {
842         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
843         spin_lock(&clp->cl_lock);
844         memcpy(&clp->cl_cb_conn, conn, sizeof(struct nfs4_cb_conn));
845         spin_unlock(&clp->cl_lock);
846 }
847
848 /*
849  * There's currently a single callback channel slot.
850  * If the slot is available, then mark it busy.  Otherwise, set the
851  * thread for sleeping on the callback RPC wait queue.
852  */
853 static bool nfsd41_cb_get_slot(struct nfs4_client *clp, struct rpc_task *task)
854 {
855         if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
856                 rpc_sleep_on(&clp->cl_cb_waitq, task, NULL);
857                 /* Race breaker */
858                 if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
859                         dprintk("%s slot is busy\n", __func__);
860                         return false;
861                 }
862                 rpc_wake_up_queued_task(&clp->cl_cb_waitq, task);
863         }
864         return true;
865 }
866
867 /*
868  * TODO: cb_sequence should support referring call lists, cachethis, multiple
869  * slots, and mark callback channel down on communication errors.
870  */
871 static void nfsd4_cb_prepare(struct rpc_task *task, void *calldata)
872 {
873         struct nfsd4_callback *cb = calldata;
874         struct nfs4_client *clp = cb->cb_clp;
875         u32 minorversion = clp->cl_minorversion;
876
877         cb->cb_minorversion = minorversion;
878         if (minorversion) {
879                 if (!nfsd41_cb_get_slot(clp, task))
880                         return;
881         }
882         rpc_call_start(task);
883 }
884
885 static void nfsd4_cb_done(struct rpc_task *task, void *calldata)
886 {
887         struct nfsd4_callback *cb = calldata;
888         struct nfs4_client *clp = cb->cb_clp;
889
890         dprintk("%s: minorversion=%d\n", __func__,
891                 clp->cl_minorversion);
892
893         if (clp->cl_minorversion) {
894                 /* No need for lock, access serialized in nfsd4_cb_prepare */
895                 if (!task->tk_status)
896                         ++clp->cl_cb_session->se_cb_seq_nr;
897                 clear_bit(0, &clp->cl_cb_slot_busy);
898                 rpc_wake_up_next(&clp->cl_cb_waitq);
899                 dprintk("%s: freed slot, new seqid=%d\n", __func__,
900                         clp->cl_cb_session->se_cb_seq_nr);
901         }
902
903         /*
904          * If the backchannel connection was shut down while this
905          * task was queued, we need to resubmit it after setting up
906          * a new backchannel connection.
907          *
908          * Note that if we lost our callback connection permanently
909          * the submission code will error out, so we don't need to
910          * handle that case here.
911          */
912         if (task->tk_flags & RPC_TASK_KILLED) {
913                 task->tk_status = 0;
914                 cb->cb_need_restart = true;
915                 return;
916         }
917
918         if (cb->cb_status) {
919                 WARN_ON_ONCE(task->tk_status);
920                 task->tk_status = cb->cb_status;
921         }
922
923         switch (cb->cb_ops->done(cb, task)) {
924         case 0:
925                 task->tk_status = 0;
926                 rpc_restart_call_prepare(task);
927                 return;
928         case 1:
929                 break;
930         case -1:
931                 /* Network partition? */
932                 nfsd4_mark_cb_down(clp, task->tk_status);
933                 break;
934         default:
935                 BUG();
936         }
937 }
938
939 static void nfsd4_cb_release(void *calldata)
940 {
941         struct nfsd4_callback *cb = calldata;
942
943         if (cb->cb_need_restart)
944                 nfsd4_run_cb(cb);
945         else
946                 cb->cb_ops->release(cb);
947
948 }
949
950 static const struct rpc_call_ops nfsd4_cb_ops = {
951         .rpc_call_prepare = nfsd4_cb_prepare,
952         .rpc_call_done = nfsd4_cb_done,
953         .rpc_release = nfsd4_cb_release,
954 };
955
956 int nfsd4_create_callback_queue(void)
957 {
958         callback_wq = create_singlethread_workqueue("nfsd4_callbacks");
959         if (!callback_wq)
960                 return -ENOMEM;
961         return 0;
962 }
963
964 void nfsd4_destroy_callback_queue(void)
965 {
966         destroy_workqueue(callback_wq);
967 }
968
969 /* must be called under the state lock */
970 void nfsd4_shutdown_callback(struct nfs4_client *clp)
971 {
972         set_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags);
973         /*
974          * Note this won't actually result in a null callback;
975          * instead, nfsd4_run_cb_null() will detect the killed
976          * client, destroy the rpc client, and stop:
977          */
978         nfsd4_run_cb(&clp->cl_cb_null);
979         flush_workqueue(callback_wq);
980 }
981
982 /* requires cl_lock: */
983 static struct nfsd4_conn * __nfsd4_find_backchannel(struct nfs4_client *clp)
984 {
985         struct nfsd4_session *s;
986         struct nfsd4_conn *c;
987
988         list_for_each_entry(s, &clp->cl_sessions, se_perclnt) {
989                 list_for_each_entry(c, &s->se_conns, cn_persession) {
990                         if (c->cn_flags & NFS4_CDFC4_BACK)
991                                 return c;
992                 }
993         }
994         return NULL;
995 }
996
997 static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
998 {
999         struct nfs4_cb_conn conn;
1000         struct nfs4_client *clp = cb->cb_clp;
1001         struct nfsd4_session *ses = NULL;
1002         struct nfsd4_conn *c;
1003         int err;
1004
1005         /*
1006          * This is either an update, or the client dying; in either case,
1007          * kill the old client:
1008          */
1009         if (clp->cl_cb_client) {
1010                 rpc_shutdown_client(clp->cl_cb_client);
1011                 clp->cl_cb_client = NULL;
1012                 put_rpccred(clp->cl_cb_cred);
1013                 clp->cl_cb_cred = NULL;
1014         }
1015         if (clp->cl_cb_conn.cb_xprt) {
1016                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1017                 clp->cl_cb_conn.cb_xprt = NULL;
1018         }
1019         if (test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags))
1020                 return;
1021         spin_lock(&clp->cl_lock);
1022         /*
1023          * Only serialized callback code is allowed to clear these
1024          * flags; main nfsd code can only set them:
1025          */
1026         BUG_ON(!(clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK));
1027         clear_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
1028         memcpy(&conn, &cb->cb_clp->cl_cb_conn, sizeof(struct nfs4_cb_conn));
1029         c = __nfsd4_find_backchannel(clp);
1030         if (c) {
1031                 svc_xprt_get(c->cn_xprt);
1032                 conn.cb_xprt = c->cn_xprt;
1033                 ses = c->cn_session;
1034         }
1035         spin_unlock(&clp->cl_lock);
1036
1037         err = setup_callback_client(clp, &conn, ses);
1038         if (err) {
1039                 nfsd4_mark_cb_down(clp, err);
1040                 return;
1041         }
1042 }
1043
1044 static void
1045 nfsd4_run_cb_work(struct work_struct *work)
1046 {
1047         struct nfsd4_callback *cb =
1048                 container_of(work, struct nfsd4_callback, cb_work);
1049         struct nfs4_client *clp = cb->cb_clp;
1050         struct rpc_clnt *clnt;
1051
1052         if (cb->cb_need_restart) {
1053                 cb->cb_need_restart = false;
1054         } else {
1055                 if (cb->cb_ops && cb->cb_ops->prepare)
1056                         cb->cb_ops->prepare(cb);
1057         }
1058
1059         if (clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK)
1060                 nfsd4_process_cb_update(cb);
1061
1062         clnt = clp->cl_cb_client;
1063         if (!clnt) {
1064                 /* Callback channel broken, or client killed; give up: */
1065                 if (cb->cb_ops && cb->cb_ops->release)
1066                         cb->cb_ops->release(cb);
1067                 return;
1068         }
1069
1070         /*
1071          * Don't send probe messages for 4.1 or later.
1072          */
1073         if (!cb->cb_ops && clp->cl_minorversion) {
1074                 clp->cl_cb_state = NFSD4_CB_UP;
1075                 return;
1076         }
1077
1078         cb->cb_msg.rpc_cred = clp->cl_cb_cred;
1079         rpc_call_async(clnt, &cb->cb_msg, RPC_TASK_SOFT | RPC_TASK_SOFTCONN,
1080                         cb->cb_ops ? &nfsd4_cb_ops : &nfsd4_cb_probe_ops, cb);
1081 }
1082
1083 void nfsd4_init_cb(struct nfsd4_callback *cb, struct nfs4_client *clp,
1084                 struct nfsd4_callback_ops *ops, enum nfsd4_cb_op op)
1085 {
1086         cb->cb_clp = clp;
1087         cb->cb_msg.rpc_proc = &nfs4_cb_procedures[op];
1088         cb->cb_msg.rpc_argp = cb;
1089         cb->cb_msg.rpc_resp = cb;
1090         cb->cb_ops = ops;
1091         INIT_WORK(&cb->cb_work, nfsd4_run_cb_work);
1092         cb->cb_status = 0;
1093         cb->cb_need_restart = false;
1094 }
1095
1096 void nfsd4_run_cb(struct nfsd4_callback *cb)
1097 {
1098         queue_work(callback_wq, &cb->cb_work);
1099 }