GNU Linux-libre 4.19.211-gnu1
[releases.git] / 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,
227                                enum nfs_cb_opnum4 expected, int *status)
228 {
229         __be32 *p;
230         u32 op;
231
232         p = xdr_inline_decode(xdr, 4 + 4);
233         if (unlikely(p == NULL))
234                 goto out_overflow;
235         op = be32_to_cpup(p++);
236         if (unlikely(op != expected))
237                 goto out_unexpected;
238         *status = nfs_cb_stat_to_errno(be32_to_cpup(p));
239         return 0;
240 out_overflow:
241         print_overflow_msg(__func__, xdr);
242         return -EIO;
243 out_unexpected:
244         dprintk("NFSD: Callback server returned operation %d but "
245                 "we issued a request for %d\n", op, expected);
246         return -EIO;
247 }
248
249 /*
250  * CB_COMPOUND4args
251  *
252  *      struct CB_COMPOUND4args {
253  *              utf8str_cs      tag;
254  *              uint32_t        minorversion;
255  *              uint32_t        callback_ident;
256  *              nfs_cb_argop4   argarray<>;
257  *      };
258 */
259 static void encode_cb_compound4args(struct xdr_stream *xdr,
260                                     struct nfs4_cb_compound_hdr *hdr)
261 {
262         __be32 * p;
263
264         p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4);
265         p = xdr_encode_empty_array(p);          /* empty tag */
266         *p++ = cpu_to_be32(hdr->minorversion);
267         *p++ = cpu_to_be32(hdr->ident);
268
269         hdr->nops_p = p;
270         *p = cpu_to_be32(hdr->nops);            /* argarray element count */
271 }
272
273 /*
274  * Update argarray element count
275  */
276 static void encode_cb_nops(struct nfs4_cb_compound_hdr *hdr)
277 {
278         BUG_ON(hdr->nops > NFS4_MAX_BACK_CHANNEL_OPS);
279         *hdr->nops_p = cpu_to_be32(hdr->nops);
280 }
281
282 /*
283  * CB_COMPOUND4res
284  *
285  *      struct CB_COMPOUND4res {
286  *              nfsstat4        status;
287  *              utf8str_cs      tag;
288  *              nfs_cb_resop4   resarray<>;
289  *      };
290  */
291 static int decode_cb_compound4res(struct xdr_stream *xdr,
292                                   struct nfs4_cb_compound_hdr *hdr)
293 {
294         u32 length;
295         __be32 *p;
296
297         p = xdr_inline_decode(xdr, 4 + 4);
298         if (unlikely(p == NULL))
299                 goto out_overflow;
300         hdr->status = be32_to_cpup(p++);
301         /* Ignore the tag */
302         length = be32_to_cpup(p++);
303         p = xdr_inline_decode(xdr, length + 4);
304         if (unlikely(p == NULL))
305                 goto out_overflow;
306         p += XDR_QUADLEN(length);
307         hdr->nops = be32_to_cpup(p);
308         return 0;
309 out_overflow:
310         print_overflow_msg(__func__, xdr);
311         return -EIO;
312 }
313
314 /*
315  * CB_RECALL4args
316  *
317  *      struct CB_RECALL4args {
318  *              stateid4        stateid;
319  *              bool            truncate;
320  *              nfs_fh4         fh;
321  *      };
322  */
323 static void encode_cb_recall4args(struct xdr_stream *xdr,
324                                   const struct nfs4_delegation *dp,
325                                   struct nfs4_cb_compound_hdr *hdr)
326 {
327         __be32 *p;
328
329         encode_nfs_cb_opnum4(xdr, OP_CB_RECALL);
330         encode_stateid4(xdr, &dp->dl_stid.sc_stateid);
331
332         p = xdr_reserve_space(xdr, 4);
333         *p++ = xdr_zero;                        /* truncate */
334
335         encode_nfs_fh4(xdr, &dp->dl_stid.sc_file->fi_fhandle);
336
337         hdr->nops++;
338 }
339
340 /*
341  * CB_SEQUENCE4args
342  *
343  *      struct CB_SEQUENCE4args {
344  *              sessionid4              csa_sessionid;
345  *              sequenceid4             csa_sequenceid;
346  *              slotid4                 csa_slotid;
347  *              slotid4                 csa_highest_slotid;
348  *              bool                    csa_cachethis;
349  *              referring_call_list4    csa_referring_call_lists<>;
350  *      };
351  */
352 static void encode_cb_sequence4args(struct xdr_stream *xdr,
353                                     const struct nfsd4_callback *cb,
354                                     struct nfs4_cb_compound_hdr *hdr)
355 {
356         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
357         __be32 *p;
358
359         if (hdr->minorversion == 0)
360                 return;
361
362         encode_nfs_cb_opnum4(xdr, OP_CB_SEQUENCE);
363         encode_sessionid4(xdr, session);
364
365         p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4 + 4);
366         *p++ = cpu_to_be32(session->se_cb_seq_nr);      /* csa_sequenceid */
367         *p++ = xdr_zero;                        /* csa_slotid */
368         *p++ = xdr_zero;                        /* csa_highest_slotid */
369         *p++ = xdr_zero;                        /* csa_cachethis */
370         xdr_encode_empty_array(p);              /* csa_referring_call_lists */
371
372         hdr->nops++;
373 }
374
375 /*
376  * CB_SEQUENCE4resok
377  *
378  *      struct CB_SEQUENCE4resok {
379  *              sessionid4      csr_sessionid;
380  *              sequenceid4     csr_sequenceid;
381  *              slotid4         csr_slotid;
382  *              slotid4         csr_highest_slotid;
383  *              slotid4         csr_target_highest_slotid;
384  *      };
385  *
386  *      union CB_SEQUENCE4res switch (nfsstat4 csr_status) {
387  *      case NFS4_OK:
388  *              CB_SEQUENCE4resok       csr_resok4;
389  *      default:
390  *              void;
391  *      };
392  *
393  * Our current back channel implmentation supports a single backchannel
394  * with a single slot.
395  */
396 static int decode_cb_sequence4resok(struct xdr_stream *xdr,
397                                     struct nfsd4_callback *cb)
398 {
399         struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
400         int status = -ESERVERFAULT;
401         __be32 *p;
402         u32 dummy;
403
404         /*
405          * If the server returns different values for sessionID, slotID or
406          * sequence number, the server is looney tunes.
407          */
408         p = xdr_inline_decode(xdr, NFS4_MAX_SESSIONID_LEN + 4 + 4 + 4 + 4);
409         if (unlikely(p == NULL))
410                 goto out_overflow;
411
412         if (memcmp(p, session->se_sessionid.data, NFS4_MAX_SESSIONID_LEN)) {
413                 dprintk("NFS: %s Invalid session id\n", __func__);
414                 goto out;
415         }
416         p += XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN);
417
418         dummy = be32_to_cpup(p++);
419         if (dummy != session->se_cb_seq_nr) {
420                 dprintk("NFS: %s Invalid sequence number\n", __func__);
421                 goto out;
422         }
423
424         dummy = be32_to_cpup(p++);
425         if (dummy != 0) {
426                 dprintk("NFS: %s Invalid slotid\n", __func__);
427                 goto out;
428         }
429
430         /*
431          * FIXME: process highest slotid and target highest slotid
432          */
433         status = 0;
434 out:
435         cb->cb_seq_status = status;
436         return status;
437 out_overflow:
438         print_overflow_msg(__func__, xdr);
439         status = -EIO;
440         goto out;
441 }
442
443 static int decode_cb_sequence4res(struct xdr_stream *xdr,
444                                   struct nfsd4_callback *cb)
445 {
446         int status;
447
448         if (cb->cb_clp->cl_minorversion == 0)
449                 return 0;
450
451         status = decode_cb_op_status(xdr, OP_CB_SEQUENCE, &cb->cb_seq_status);
452         if (unlikely(status || cb->cb_seq_status))
453                 return status;
454
455         return decode_cb_sequence4resok(xdr, cb);
456 }
457
458 /*
459  * NFSv4.0 and NFSv4.1 XDR encode functions
460  *
461  * NFSv4.0 callback argument types are defined in section 15 of RFC
462  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
463  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
464  * Protocol".
465  */
466
467 /*
468  * NB: Without this zero space reservation, callbacks over krb5p fail
469  */
470 static void nfs4_xdr_enc_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
471                                  const void *__unused)
472 {
473         xdr_reserve_space(xdr, 0);
474 }
475
476 /*
477  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
478  */
479 static void nfs4_xdr_enc_cb_recall(struct rpc_rqst *req, struct xdr_stream *xdr,
480                                    const void *data)
481 {
482         const struct nfsd4_callback *cb = data;
483         const struct nfs4_delegation *dp = cb_to_delegation(cb);
484         struct nfs4_cb_compound_hdr hdr = {
485                 .ident = cb->cb_clp->cl_cb_ident,
486                 .minorversion = cb->cb_clp->cl_minorversion,
487         };
488
489         encode_cb_compound4args(xdr, &hdr);
490         encode_cb_sequence4args(xdr, cb, &hdr);
491         encode_cb_recall4args(xdr, dp, &hdr);
492         encode_cb_nops(&hdr);
493 }
494
495
496 /*
497  * NFSv4.0 and NFSv4.1 XDR decode functions
498  *
499  * NFSv4.0 callback result types are defined in section 15 of RFC
500  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
501  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
502  * Protocol".
503  */
504
505 static int nfs4_xdr_dec_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
506                                 void *__unused)
507 {
508         return 0;
509 }
510
511 /*
512  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
513  */
514 static int nfs4_xdr_dec_cb_recall(struct rpc_rqst *rqstp,
515                                   struct xdr_stream *xdr,
516                                   void *data)
517 {
518         struct nfsd4_callback *cb = data;
519         struct nfs4_cb_compound_hdr hdr;
520         int status;
521
522         status = decode_cb_compound4res(xdr, &hdr);
523         if (unlikely(status))
524                 return status;
525
526         if (cb != NULL) {
527                 status = decode_cb_sequence4res(xdr, cb);
528                 if (unlikely(status || cb->cb_seq_status))
529                         return status;
530         }
531
532         return decode_cb_op_status(xdr, OP_CB_RECALL, &cb->cb_status);
533 }
534
535 #ifdef CONFIG_NFSD_PNFS
536 /*
537  * CB_LAYOUTRECALL4args
538  *
539  *      struct layoutrecall_file4 {
540  *              nfs_fh4         lor_fh;
541  *              offset4         lor_offset;
542  *              length4         lor_length;
543  *              stateid4        lor_stateid;
544  *      };
545  *
546  *      union layoutrecall4 switch(layoutrecall_type4 lor_recalltype) {
547  *      case LAYOUTRECALL4_FILE:
548  *              layoutrecall_file4 lor_layout;
549  *      case LAYOUTRECALL4_FSID:
550  *              fsid4              lor_fsid;
551  *      case LAYOUTRECALL4_ALL:
552  *              void;
553  *      };
554  *
555  *      struct CB_LAYOUTRECALL4args {
556  *              layouttype4             clora_type;
557  *              layoutiomode4           clora_iomode;
558  *              bool                    clora_changed;
559  *              layoutrecall4           clora_recall;
560  *      };
561  */
562 static void encode_cb_layout4args(struct xdr_stream *xdr,
563                                   const struct nfs4_layout_stateid *ls,
564                                   struct nfs4_cb_compound_hdr *hdr)
565 {
566         __be32 *p;
567
568         BUG_ON(hdr->minorversion == 0);
569
570         p = xdr_reserve_space(xdr, 5 * 4);
571         *p++ = cpu_to_be32(OP_CB_LAYOUTRECALL);
572         *p++ = cpu_to_be32(ls->ls_layout_type);
573         *p++ = cpu_to_be32(IOMODE_ANY);
574         *p++ = cpu_to_be32(1);
575         *p = cpu_to_be32(RETURN_FILE);
576
577         encode_nfs_fh4(xdr, &ls->ls_stid.sc_file->fi_fhandle);
578
579         p = xdr_reserve_space(xdr, 2 * 8);
580         p = xdr_encode_hyper(p, 0);
581         xdr_encode_hyper(p, NFS4_MAX_UINT64);
582
583         encode_stateid4(xdr, &ls->ls_recall_sid);
584
585         hdr->nops++;
586 }
587
588 static void nfs4_xdr_enc_cb_layout(struct rpc_rqst *req,
589                                    struct xdr_stream *xdr,
590                                    const void *data)
591 {
592         const struct nfsd4_callback *cb = data;
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_clp->cl_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                                   void *data)
609 {
610         struct nfsd4_callback *cb = data;
611         struct nfs4_cb_compound_hdr hdr;
612         int status;
613
614         status = decode_cb_compound4res(xdr, &hdr);
615         if (unlikely(status))
616                 return status;
617
618         if (cb) {
619                 status = decode_cb_sequence4res(xdr, cb);
620                 if (unlikely(status || cb->cb_seq_status))
621                         return status;
622         }
623         return decode_cb_op_status(xdr, OP_CB_LAYOUTRECALL, &cb->cb_status);
624 }
625 #endif /* CONFIG_NFSD_PNFS */
626
627 static void encode_stateowner(struct xdr_stream *xdr, struct nfs4_stateowner *so)
628 {
629         __be32  *p;
630
631         p = xdr_reserve_space(xdr, 8 + 4 + so->so_owner.len);
632         p = xdr_encode_opaque_fixed(p, &so->so_client->cl_clientid, 8);
633         xdr_encode_opaque(p, so->so_owner.data, so->so_owner.len);
634 }
635
636 static void nfs4_xdr_enc_cb_notify_lock(struct rpc_rqst *req,
637                                         struct xdr_stream *xdr,
638                                         const void *data)
639 {
640         const struct nfsd4_callback *cb = data;
641         const struct nfsd4_blocked_lock *nbl =
642                 container_of(cb, struct nfsd4_blocked_lock, nbl_cb);
643         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)nbl->nbl_lock.fl_owner;
644         struct nfs4_cb_compound_hdr hdr = {
645                 .ident = 0,
646                 .minorversion = cb->cb_clp->cl_minorversion,
647         };
648
649         __be32 *p;
650
651         BUG_ON(hdr.minorversion == 0);
652
653         encode_cb_compound4args(xdr, &hdr);
654         encode_cb_sequence4args(xdr, cb, &hdr);
655
656         p = xdr_reserve_space(xdr, 4);
657         *p = cpu_to_be32(OP_CB_NOTIFY_LOCK);
658         encode_nfs_fh4(xdr, &nbl->nbl_fh);
659         encode_stateowner(xdr, &lo->lo_owner);
660         hdr.nops++;
661
662         encode_cb_nops(&hdr);
663 }
664
665 static int nfs4_xdr_dec_cb_notify_lock(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         if (cb) {
678                 status = decode_cb_sequence4res(xdr, cb);
679                 if (unlikely(status || cb->cb_seq_status))
680                         return status;
681         }
682         return decode_cb_op_status(xdr, OP_CB_NOTIFY_LOCK, &cb->cb_status);
683 }
684
685 /*
686  * RPC procedure tables
687  */
688 #define PROC(proc, call, argtype, restype)                              \
689 [NFSPROC4_CLNT_##proc] = {                                              \
690         .p_proc    = NFSPROC4_CB_##call,                                \
691         .p_encode  = nfs4_xdr_enc_##argtype,            \
692         .p_decode  = nfs4_xdr_dec_##restype,                            \
693         .p_arglen  = NFS4_enc_##argtype##_sz,                           \
694         .p_replen  = NFS4_dec_##restype##_sz,                           \
695         .p_statidx = NFSPROC4_CB_##call,                                \
696         .p_name    = #proc,                                             \
697 }
698
699 static const struct rpc_procinfo nfs4_cb_procedures[] = {
700         PROC(CB_NULL,   NULL,           cb_null,        cb_null),
701         PROC(CB_RECALL, COMPOUND,       cb_recall,      cb_recall),
702 #ifdef CONFIG_NFSD_PNFS
703         PROC(CB_LAYOUT, COMPOUND,       cb_layout,      cb_layout),
704 #endif
705         PROC(CB_NOTIFY_LOCK,    COMPOUND,       cb_notify_lock, cb_notify_lock),
706 };
707
708 static unsigned int nfs4_cb_counts[ARRAY_SIZE(nfs4_cb_procedures)];
709 static const struct rpc_version nfs_cb_version4 = {
710 /*
711  * Note on the callback rpc program version number: despite language in rfc
712  * 5661 section 18.36.3 requiring servers to use 4 in this field, the
713  * official xdr descriptions for both 4.0 and 4.1 specify version 1, and
714  * in practice that appears to be what implementations use.  The section
715  * 18.36.3 language is expected to be fixed in an erratum.
716  */
717         .number                 = 1,
718         .nrprocs                = ARRAY_SIZE(nfs4_cb_procedures),
719         .procs                  = nfs4_cb_procedures,
720         .counts                 = nfs4_cb_counts,
721 };
722
723 static const struct rpc_version *nfs_cb_version[2] = {
724         [1] = &nfs_cb_version4,
725 };
726
727 static const struct rpc_program cb_program;
728
729 static struct rpc_stat cb_stats = {
730         .program                = &cb_program
731 };
732
733 #define NFS4_CALLBACK 0x40000000
734 static const struct rpc_program cb_program = {
735         .name                   = "nfs4_cb",
736         .number                 = NFS4_CALLBACK,
737         .nrvers                 = ARRAY_SIZE(nfs_cb_version),
738         .version                = nfs_cb_version,
739         .stats                  = &cb_stats,
740         .pipe_dir_name          = "nfsd4_cb",
741 };
742
743 static int max_cb_time(struct net *net)
744 {
745         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
746         return max(nn->nfsd4_lease/10, (time_t)1) * HZ;
747 }
748
749 static struct rpc_cred *get_backchannel_cred(struct nfs4_client *clp, struct rpc_clnt *client, struct nfsd4_session *ses)
750 {
751         if (clp->cl_minorversion == 0) {
752                 char *principal = clp->cl_cred.cr_targ_princ ?
753                                         clp->cl_cred.cr_targ_princ : "nfs";
754                 struct rpc_cred *cred;
755
756                 cred = rpc_lookup_machine_cred(principal);
757                 if (!IS_ERR(cred))
758                         get_rpccred(cred);
759                 return cred;
760         } else {
761                 struct rpc_auth *auth = client->cl_auth;
762                 struct auth_cred acred = {};
763
764                 acred.uid = ses->se_cb_sec.uid;
765                 acred.gid = ses->se_cb_sec.gid;
766                 return auth->au_ops->lookup_cred(client->cl_auth, &acred, 0);
767         }
768 }
769
770 static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn, struct nfsd4_session *ses)
771 {
772         int maxtime = max_cb_time(clp->net);
773         struct rpc_timeout      timeparms = {
774                 .to_initval     = maxtime,
775                 .to_retries     = 0,
776                 .to_maxval      = maxtime,
777         };
778         struct rpc_create_args args = {
779                 .net            = clp->net,
780                 .address        = (struct sockaddr *) &conn->cb_addr,
781                 .addrsize       = conn->cb_addrlen,
782                 .saddress       = (struct sockaddr *) &conn->cb_saddr,
783                 .timeout        = &timeparms,
784                 .program        = &cb_program,
785                 .version        = 1,
786                 .flags          = (RPC_CLNT_CREATE_NOPING | RPC_CLNT_CREATE_QUIET),
787         };
788         struct rpc_clnt *client;
789         struct rpc_cred *cred;
790
791         if (clp->cl_minorversion == 0) {
792                 if (!clp->cl_cred.cr_principal &&
793                                 (clp->cl_cred.cr_flavor >= RPC_AUTH_GSS_KRB5))
794                         return -EINVAL;
795                 args.client_name = clp->cl_cred.cr_principal;
796                 args.prognumber = conn->cb_prog;
797                 args.protocol = XPRT_TRANSPORT_TCP;
798                 args.authflavor = clp->cl_cred.cr_flavor;
799                 clp->cl_cb_ident = conn->cb_ident;
800         } else {
801                 if (!conn->cb_xprt)
802                         return -EINVAL;
803                 clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
804                 clp->cl_cb_session = ses;
805                 args.bc_xprt = conn->cb_xprt;
806                 args.prognumber = clp->cl_cb_session->se_cb_prog;
807                 args.protocol = conn->cb_xprt->xpt_class->xcl_ident |
808                                 XPRT_TRANSPORT_BC;
809                 args.authflavor = ses->se_cb_sec.flavor;
810         }
811         /* Create RPC client */
812         client = rpc_create(&args);
813         if (IS_ERR(client)) {
814                 dprintk("NFSD: couldn't create callback client: %ld\n",
815                         PTR_ERR(client));
816                 return PTR_ERR(client);
817         }
818         cred = get_backchannel_cred(clp, client, ses);
819         if (IS_ERR(cred)) {
820                 rpc_shutdown_client(client);
821                 return PTR_ERR(cred);
822         }
823         clp->cl_cb_client = client;
824         clp->cl_cb_cred = cred;
825         return 0;
826 }
827
828 static void warn_no_callback_path(struct nfs4_client *clp, int reason)
829 {
830         dprintk("NFSD: warning: no callback path to client %.*s: error %d\n",
831                 (int)clp->cl_name.len, clp->cl_name.data, reason);
832 }
833
834 static void nfsd4_mark_cb_down(struct nfs4_client *clp, int reason)
835 {
836         if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
837                 return;
838         clp->cl_cb_state = NFSD4_CB_DOWN;
839         warn_no_callback_path(clp, reason);
840 }
841
842 static void nfsd4_mark_cb_fault(struct nfs4_client *clp, int reason)
843 {
844         if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
845                 return;
846         clp->cl_cb_state = NFSD4_CB_FAULT;
847         warn_no_callback_path(clp, reason);
848 }
849
850 static void nfsd4_cb_probe_done(struct rpc_task *task, void *calldata)
851 {
852         struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
853
854         if (task->tk_status)
855                 nfsd4_mark_cb_down(clp, task->tk_status);
856         else
857                 clp->cl_cb_state = NFSD4_CB_UP;
858 }
859
860 static const struct rpc_call_ops nfsd4_cb_probe_ops = {
861         /* XXX: release method to ensure we set the cb channel down if
862          * necessary on early failure? */
863         .rpc_call_done = nfsd4_cb_probe_done,
864 };
865
866 static struct workqueue_struct *callback_wq;
867
868 /*
869  * Poke the callback thread to process any updates to the callback
870  * parameters, and send a null probe.
871  */
872 void nfsd4_probe_callback(struct nfs4_client *clp)
873 {
874         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
875         set_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
876         nfsd4_run_cb(&clp->cl_cb_null);
877 }
878
879 void nfsd4_probe_callback_sync(struct nfs4_client *clp)
880 {
881         nfsd4_probe_callback(clp);
882         flush_workqueue(callback_wq);
883 }
884
885 void nfsd4_change_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
886 {
887         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
888         spin_lock(&clp->cl_lock);
889         memcpy(&clp->cl_cb_conn, conn, sizeof(struct nfs4_cb_conn));
890         spin_unlock(&clp->cl_lock);
891 }
892
893 /*
894  * There's currently a single callback channel slot.
895  * If the slot is available, then mark it busy.  Otherwise, set the
896  * thread for sleeping on the callback RPC wait queue.
897  */
898 static bool nfsd41_cb_get_slot(struct nfs4_client *clp, struct rpc_task *task)
899 {
900         if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
901                 rpc_sleep_on(&clp->cl_cb_waitq, task, NULL);
902                 /* Race breaker */
903                 if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
904                         dprintk("%s slot is busy\n", __func__);
905                         return false;
906                 }
907                 rpc_wake_up_queued_task(&clp->cl_cb_waitq, task);
908         }
909         return true;
910 }
911
912 /*
913  * TODO: cb_sequence should support referring call lists, cachethis, multiple
914  * slots, and mark callback channel down on communication errors.
915  */
916 static void nfsd4_cb_prepare(struct rpc_task *task, void *calldata)
917 {
918         struct nfsd4_callback *cb = calldata;
919         struct nfs4_client *clp = cb->cb_clp;
920         u32 minorversion = clp->cl_minorversion;
921
922         /*
923          * cb_seq_status is only set in decode_cb_sequence4res,
924          * and so will remain 1 if an rpc level failure occurs.
925          */
926         cb->cb_seq_status = 1;
927         cb->cb_status = 0;
928         if (minorversion) {
929                 if (!cb->cb_holds_slot && !nfsd41_cb_get_slot(clp, task))
930                         return;
931                 cb->cb_holds_slot = true;
932         }
933         rpc_call_start(task);
934 }
935
936 static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback *cb)
937 {
938         struct nfs4_client *clp = cb->cb_clp;
939         struct nfsd4_session *session = clp->cl_cb_session;
940         bool ret = true;
941
942         if (!clp->cl_minorversion) {
943                 /*
944                  * If the backchannel connection was shut down while this
945                  * task was queued, we need to resubmit it after setting up
946                  * a new backchannel connection.
947                  *
948                  * Note that if we lost our callback connection permanently
949                  * the submission code will error out, so we don't need to
950                  * handle that case here.
951                  */
952                 if (task->tk_flags & RPC_TASK_KILLED)
953                         goto need_restart;
954
955                 return true;
956         }
957
958         if (!cb->cb_holds_slot)
959                 goto need_restart;
960
961         switch (cb->cb_seq_status) {
962         case 0:
963                 /*
964                  * No need for lock, access serialized in nfsd4_cb_prepare
965                  *
966                  * RFC5661 20.9.3
967                  * If CB_SEQUENCE returns an error, then the state of the slot
968                  * (sequence ID, cached reply) MUST NOT change.
969                  */
970                 ++session->se_cb_seq_nr;
971                 break;
972         case -ESERVERFAULT:
973                 ++session->se_cb_seq_nr;
974                 /* Fall through */
975         case 1:
976         case -NFS4ERR_BADSESSION:
977                 nfsd4_mark_cb_fault(cb->cb_clp, cb->cb_seq_status);
978                 ret = false;
979                 break;
980         case -NFS4ERR_DELAY:
981                 if (!rpc_restart_call(task))
982                         goto out;
983
984                 rpc_delay(task, 2 * HZ);
985                 return false;
986         case -NFS4ERR_BADSLOT:
987                 goto retry_nowait;
988         case -NFS4ERR_SEQ_MISORDERED:
989                 if (session->se_cb_seq_nr != 1) {
990                         session->se_cb_seq_nr = 1;
991                         goto retry_nowait;
992                 }
993                 break;
994         default:
995                 dprintk("%s: unprocessed error %d\n", __func__,
996                         cb->cb_seq_status);
997         }
998
999         cb->cb_holds_slot = false;
1000         clear_bit(0, &clp->cl_cb_slot_busy);
1001         rpc_wake_up_next(&clp->cl_cb_waitq);
1002         dprintk("%s: freed slot, new seqid=%d\n", __func__,
1003                 clp->cl_cb_session->se_cb_seq_nr);
1004
1005         if (task->tk_flags & RPC_TASK_KILLED)
1006                 goto need_restart;
1007 out:
1008         return ret;
1009 retry_nowait:
1010         if (rpc_restart_call_prepare(task))
1011                 ret = false;
1012         goto out;
1013 need_restart:
1014         task->tk_status = 0;
1015         cb->cb_need_restart = true;
1016         return false;
1017 }
1018
1019 static void nfsd4_cb_done(struct rpc_task *task, void *calldata)
1020 {
1021         struct nfsd4_callback *cb = calldata;
1022         struct nfs4_client *clp = cb->cb_clp;
1023
1024         dprintk("%s: minorversion=%d\n", __func__,
1025                 clp->cl_minorversion);
1026
1027         if (!nfsd4_cb_sequence_done(task, cb))
1028                 return;
1029
1030         if (cb->cb_status) {
1031                 WARN_ON_ONCE(task->tk_status);
1032                 task->tk_status = cb->cb_status;
1033         }
1034
1035         switch (cb->cb_ops->done(cb, task)) {
1036         case 0:
1037                 task->tk_status = 0;
1038                 rpc_restart_call_prepare(task);
1039                 return;
1040         case 1:
1041                 break;
1042         case -1:
1043                 /* Network partition? */
1044                 nfsd4_mark_cb_down(clp, task->tk_status);
1045                 break;
1046         default:
1047                 BUG();
1048         }
1049 }
1050
1051 static void nfsd4_cb_release(void *calldata)
1052 {
1053         struct nfsd4_callback *cb = calldata;
1054
1055         if (cb->cb_need_restart)
1056                 nfsd4_run_cb(cb);
1057         else
1058                 cb->cb_ops->release(cb);
1059
1060 }
1061
1062 static const struct rpc_call_ops nfsd4_cb_ops = {
1063         .rpc_call_prepare = nfsd4_cb_prepare,
1064         .rpc_call_done = nfsd4_cb_done,
1065         .rpc_release = nfsd4_cb_release,
1066 };
1067
1068 int nfsd4_create_callback_queue(void)
1069 {
1070         callback_wq = alloc_ordered_workqueue("nfsd4_callbacks", 0);
1071         if (!callback_wq)
1072                 return -ENOMEM;
1073         return 0;
1074 }
1075
1076 void nfsd4_destroy_callback_queue(void)
1077 {
1078         destroy_workqueue(callback_wq);
1079 }
1080
1081 /* must be called under the state lock */
1082 void nfsd4_shutdown_callback(struct nfs4_client *clp)
1083 {
1084         set_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags);
1085         /*
1086          * Note this won't actually result in a null callback;
1087          * instead, nfsd4_run_cb_null() will detect the killed
1088          * client, destroy the rpc client, and stop:
1089          */
1090         nfsd4_run_cb(&clp->cl_cb_null);
1091         flush_workqueue(callback_wq);
1092 }
1093
1094 /* requires cl_lock: */
1095 static struct nfsd4_conn * __nfsd4_find_backchannel(struct nfs4_client *clp)
1096 {
1097         struct nfsd4_session *s;
1098         struct nfsd4_conn *c;
1099
1100         list_for_each_entry(s, &clp->cl_sessions, se_perclnt) {
1101                 list_for_each_entry(c, &s->se_conns, cn_persession) {
1102                         if (c->cn_flags & NFS4_CDFC4_BACK)
1103                                 return c;
1104                 }
1105         }
1106         return NULL;
1107 }
1108
1109 static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
1110 {
1111         struct nfs4_cb_conn conn;
1112         struct nfs4_client *clp = cb->cb_clp;
1113         struct nfsd4_session *ses = NULL;
1114         struct nfsd4_conn *c;
1115         int err;
1116
1117         /*
1118          * This is either an update, or the client dying; in either case,
1119          * kill the old client:
1120          */
1121         if (clp->cl_cb_client) {
1122                 rpc_shutdown_client(clp->cl_cb_client);
1123                 clp->cl_cb_client = NULL;
1124                 put_rpccred(clp->cl_cb_cred);
1125                 clp->cl_cb_cred = NULL;
1126         }
1127         if (clp->cl_cb_conn.cb_xprt) {
1128                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1129                 clp->cl_cb_conn.cb_xprt = NULL;
1130         }
1131         if (test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags))
1132                 return;
1133         spin_lock(&clp->cl_lock);
1134         /*
1135          * Only serialized callback code is allowed to clear these
1136          * flags; main nfsd code can only set them:
1137          */
1138         BUG_ON(!(clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK));
1139         clear_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
1140         memcpy(&conn, &cb->cb_clp->cl_cb_conn, sizeof(struct nfs4_cb_conn));
1141         c = __nfsd4_find_backchannel(clp);
1142         if (c) {
1143                 svc_xprt_get(c->cn_xprt);
1144                 conn.cb_xprt = c->cn_xprt;
1145                 ses = c->cn_session;
1146         }
1147         spin_unlock(&clp->cl_lock);
1148
1149         err = setup_callback_client(clp, &conn, ses);
1150         if (err) {
1151                 nfsd4_mark_cb_down(clp, err);
1152                 if (c)
1153                         svc_xprt_put(c->cn_xprt);
1154                 return;
1155         }
1156 }
1157
1158 static void
1159 nfsd4_run_cb_work(struct work_struct *work)
1160 {
1161         struct nfsd4_callback *cb =
1162                 container_of(work, struct nfsd4_callback, cb_work);
1163         struct nfs4_client *clp = cb->cb_clp;
1164         struct rpc_clnt *clnt;
1165
1166         if (cb->cb_need_restart) {
1167                 cb->cb_need_restart = false;
1168         } else {
1169                 if (cb->cb_ops && cb->cb_ops->prepare)
1170                         cb->cb_ops->prepare(cb);
1171         }
1172
1173         if (clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK)
1174                 nfsd4_process_cb_update(cb);
1175
1176         clnt = clp->cl_cb_client;
1177         if (!clnt) {
1178                 /* Callback channel broken, or client killed; give up: */
1179                 if (cb->cb_ops && cb->cb_ops->release)
1180                         cb->cb_ops->release(cb);
1181                 return;
1182         }
1183
1184         /*
1185          * Don't send probe messages for 4.1 or later.
1186          */
1187         if (!cb->cb_ops && clp->cl_minorversion) {
1188                 clp->cl_cb_state = NFSD4_CB_UP;
1189                 return;
1190         }
1191
1192         cb->cb_msg.rpc_cred = clp->cl_cb_cred;
1193         rpc_call_async(clnt, &cb->cb_msg, RPC_TASK_SOFT | RPC_TASK_SOFTCONN,
1194                         cb->cb_ops ? &nfsd4_cb_ops : &nfsd4_cb_probe_ops, cb);
1195 }
1196
1197 void nfsd4_init_cb(struct nfsd4_callback *cb, struct nfs4_client *clp,
1198                 const struct nfsd4_callback_ops *ops, enum nfsd4_cb_op op)
1199 {
1200         cb->cb_clp = clp;
1201         cb->cb_msg.rpc_proc = &nfs4_cb_procedures[op];
1202         cb->cb_msg.rpc_argp = cb;
1203         cb->cb_msg.rpc_resp = cb;
1204         cb->cb_ops = ops;
1205         INIT_WORK(&cb->cb_work, nfsd4_run_cb_work);
1206         cb->cb_seq_status = 1;
1207         cb->cb_status = 0;
1208         cb->cb_need_restart = false;
1209         cb->cb_holds_slot = false;
1210 }
1211
1212 void nfsd4_run_cb(struct nfsd4_callback *cb)
1213 {
1214         queue_work(callback_wq, &cb->cb_work);
1215 }