Merge branch 'for-2.6.31' of git://git.linux-nfs.org/projects/trondmy/nfs-2.6
[linux-2.6.git] / fs / nfsd / nfs4callback.c
1 /*
2  *  linux/fs/nfsd/nfs4callback.c
3  *
4  *  Copyright (c) 2001 The Regents of the University of Michigan.
5  *  All rights reserved.
6  *
7  *  Kendrick Smith <kmsmith@umich.edu>
8  *  Andy Adamson <andros@umich.edu>
9  *
10  *  Redistribution and use in source and binary forms, with or without
11  *  modification, are permitted provided that the following conditions
12  *  are met:
13  *
14  *  1. Redistributions of source code must retain the above copyright
15  *     notice, this list of conditions and the following disclaimer.
16  *  2. Redistributions in binary form must reproduce the above copyright
17  *     notice, this list of conditions and the following disclaimer in the
18  *     documentation and/or other materials provided with the distribution.
19  *  3. Neither the name of the University nor the names of its
20  *     contributors may be used to endorse or promote products derived
21  *     from this software without specific prior written permission.
22  *
23  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34  */
35
36 #include <linux/module.h>
37 #include <linux/list.h>
38 #include <linux/inet.h>
39 #include <linux/errno.h>
40 #include <linux/delay.h>
41 #include <linux/sched.h>
42 #include <linux/kthread.h>
43 #include <linux/sunrpc/xdr.h>
44 #include <linux/sunrpc/svc.h>
45 #include <linux/sunrpc/clnt.h>
46 #include <linux/nfsd/nfsd.h>
47 #include <linux/nfsd/state.h>
48 #include <linux/sunrpc/sched.h>
49 #include <linux/nfs4.h>
50
51 #define NFSDDBG_FACILITY                NFSDDBG_PROC
52
53 #define NFSPROC4_CB_NULL 0
54 #define NFSPROC4_CB_COMPOUND 1
55
56 /* Index of predefined Linux callback client operations */
57
58 enum {
59         NFSPROC4_CLNT_CB_NULL = 0,
60         NFSPROC4_CLNT_CB_RECALL,
61 };
62
63 enum nfs_cb_opnum4 {
64         OP_CB_RECALL            = 4,
65 };
66
67 #define NFS4_MAXTAGLEN          20
68
69 #define NFS4_enc_cb_null_sz             0
70 #define NFS4_dec_cb_null_sz             0
71 #define cb_compound_enc_hdr_sz          4
72 #define cb_compound_dec_hdr_sz          (3 + (NFS4_MAXTAGLEN >> 2))
73 #define op_enc_sz                       1
74 #define op_dec_sz                       2
75 #define enc_nfs4_fh_sz                  (1 + (NFS4_FHSIZE >> 2))
76 #define enc_stateid_sz                  (NFS4_STATEID_SIZE >> 2)
77 #define NFS4_enc_cb_recall_sz           (cb_compound_enc_hdr_sz +       \
78                                         1 + enc_stateid_sz +            \
79                                         enc_nfs4_fh_sz)
80
81 #define NFS4_dec_cb_recall_sz           (cb_compound_dec_hdr_sz  +      \
82                                         op_dec_sz)
83
84 /*
85 * Generic encode routines from fs/nfs/nfs4xdr.c
86 */
87 static inline __be32 *
88 xdr_writemem(__be32 *p, const void *ptr, int nbytes)
89 {
90         int tmp = XDR_QUADLEN(nbytes);
91         if (!tmp)
92                 return p;
93         p[tmp-1] = 0;
94         memcpy(p, ptr, nbytes);
95         return p + tmp;
96 }
97
98 #define WRITE32(n)               *p++ = htonl(n)
99 #define WRITEMEM(ptr,nbytes)     do {                           \
100         p = xdr_writemem(p, ptr, nbytes);                       \
101 } while (0)
102 #define RESERVE_SPACE(nbytes)   do {                            \
103         p = xdr_reserve_space(xdr, nbytes);                     \
104         if (!p) dprintk("NFSD: RESERVE_SPACE(%d) failed in function %s\n", (int) (nbytes), __func__); \
105         BUG_ON(!p);                                             \
106 } while (0)
107
108 /*
109  * Generic decode routines from fs/nfs/nfs4xdr.c
110  */
111 #define DECODE_TAIL                             \
112         status = 0;                             \
113 out:                                            \
114         return status;                          \
115 xdr_error:                                      \
116         dprintk("NFSD: xdr error! (%s:%d)\n", __FILE__, __LINE__); \
117         status = -EIO;                          \
118         goto out
119
120 #define READ32(x)         (x) = ntohl(*p++)
121 #define READ64(x)         do {                  \
122         (x) = (u64)ntohl(*p++) << 32;           \
123         (x) |= ntohl(*p++);                     \
124 } while (0)
125 #define READTIME(x)       do {                  \
126         p++;                                    \
127         (x.tv_sec) = ntohl(*p++);               \
128         (x.tv_nsec) = ntohl(*p++);              \
129 } while (0)
130 #define READ_BUF(nbytes)  do { \
131         p = xdr_inline_decode(xdr, nbytes); \
132         if (!p) { \
133                 dprintk("NFSD: %s: reply buffer overflowed in line %d.\n", \
134                         __func__, __LINE__); \
135                 return -EIO; \
136         } \
137 } while (0)
138
139 struct nfs4_cb_compound_hdr {
140         int             status;
141         u32             ident;
142         u32             nops;
143         u32             taglen;
144         char *          tag;
145 };
146
147 static struct {
148 int stat;
149 int errno;
150 } nfs_cb_errtbl[] = {
151         { NFS4_OK,              0               },
152         { NFS4ERR_PERM,         EPERM           },
153         { NFS4ERR_NOENT,        ENOENT          },
154         { NFS4ERR_IO,           EIO             },
155         { NFS4ERR_NXIO,         ENXIO           },
156         { NFS4ERR_ACCESS,       EACCES          },
157         { NFS4ERR_EXIST,        EEXIST          },
158         { NFS4ERR_XDEV,         EXDEV           },
159         { NFS4ERR_NOTDIR,       ENOTDIR         },
160         { NFS4ERR_ISDIR,        EISDIR          },
161         { NFS4ERR_INVAL,        EINVAL          },
162         { NFS4ERR_FBIG,         EFBIG           },
163         { NFS4ERR_NOSPC,        ENOSPC          },
164         { NFS4ERR_ROFS,         EROFS           },
165         { NFS4ERR_MLINK,        EMLINK          },
166         { NFS4ERR_NAMETOOLONG,  ENAMETOOLONG    },
167         { NFS4ERR_NOTEMPTY,     ENOTEMPTY       },
168         { NFS4ERR_DQUOT,        EDQUOT          },
169         { NFS4ERR_STALE,        ESTALE          },
170         { NFS4ERR_BADHANDLE,    EBADHANDLE      },
171         { NFS4ERR_BAD_COOKIE,   EBADCOOKIE      },
172         { NFS4ERR_NOTSUPP,      ENOTSUPP        },
173         { NFS4ERR_TOOSMALL,     ETOOSMALL       },
174         { NFS4ERR_SERVERFAULT,  ESERVERFAULT    },
175         { NFS4ERR_BADTYPE,      EBADTYPE        },
176         { NFS4ERR_LOCKED,       EAGAIN          },
177         { NFS4ERR_RESOURCE,     EREMOTEIO       },
178         { NFS4ERR_SYMLINK,      ELOOP           },
179         { NFS4ERR_OP_ILLEGAL,   EOPNOTSUPP      },
180         { NFS4ERR_DEADLOCK,     EDEADLK         },
181         { -1,                   EIO             }
182 };
183
184 static int
185 nfs_cb_stat_to_errno(int stat)
186 {
187         int i;
188         for (i = 0; nfs_cb_errtbl[i].stat != -1; i++) {
189                 if (nfs_cb_errtbl[i].stat == stat)
190                         return nfs_cb_errtbl[i].errno;
191         }
192         /* If we cannot translate the error, the recovery routines should
193         * handle it.
194         * Note: remaining NFSv4 error codes have values > 10000, so should
195         * not conflict with native Linux error codes.
196         */
197         return stat;
198 }
199
200 /*
201  * XDR encode
202  */
203
204 static int
205 encode_cb_compound_hdr(struct xdr_stream *xdr, struct nfs4_cb_compound_hdr *hdr)
206 {
207         __be32 * p;
208
209         RESERVE_SPACE(16);
210         WRITE32(0);            /* tag length is always 0 */
211         WRITE32(NFS4_MINOR_VERSION);
212         WRITE32(hdr->ident);
213         WRITE32(hdr->nops);
214         return 0;
215 }
216
217 static int
218 encode_cb_recall(struct xdr_stream *xdr, struct nfs4_cb_recall *cb_rec)
219 {
220         __be32 *p;
221         int len = cb_rec->cbr_fh.fh_size;
222
223         RESERVE_SPACE(12+sizeof(cb_rec->cbr_stateid) + len);
224         WRITE32(OP_CB_RECALL);
225         WRITE32(cb_rec->cbr_stateid.si_generation);
226         WRITEMEM(&cb_rec->cbr_stateid.si_opaque, sizeof(stateid_opaque_t));
227         WRITE32(cb_rec->cbr_trunc);
228         WRITE32(len);
229         WRITEMEM(&cb_rec->cbr_fh.fh_base, len);
230         return 0;
231 }
232
233 static int
234 nfs4_xdr_enc_cb_null(struct rpc_rqst *req, __be32 *p)
235 {
236         struct xdr_stream xdrs, *xdr = &xdrs;
237
238         xdr_init_encode(&xdrs, &req->rq_snd_buf, p);
239         RESERVE_SPACE(0);
240         return 0;
241 }
242
243 static int
244 nfs4_xdr_enc_cb_recall(struct rpc_rqst *req, __be32 *p, struct nfs4_cb_recall *args)
245 {
246         struct xdr_stream xdr;
247         struct nfs4_cb_compound_hdr hdr = {
248                 .ident = args->cbr_ident,
249                 .nops   = 1,
250         };
251
252         xdr_init_encode(&xdr, &req->rq_snd_buf, p);
253         encode_cb_compound_hdr(&xdr, &hdr);
254         return (encode_cb_recall(&xdr, args));
255 }
256
257
258 static int
259 decode_cb_compound_hdr(struct xdr_stream *xdr, struct nfs4_cb_compound_hdr *hdr){
260         __be32 *p;
261
262         READ_BUF(8);
263         READ32(hdr->status);
264         READ32(hdr->taglen);
265         READ_BUF(hdr->taglen + 4);
266         hdr->tag = (char *)p;
267         p += XDR_QUADLEN(hdr->taglen);
268         READ32(hdr->nops);
269         return 0;
270 }
271
272 static int
273 decode_cb_op_hdr(struct xdr_stream *xdr, enum nfs_opnum4 expected)
274 {
275         __be32 *p;
276         u32 op;
277         int32_t nfserr;
278
279         READ_BUF(8);
280         READ32(op);
281         if (op != expected) {
282                 dprintk("NFSD: decode_cb_op_hdr: Callback server returned "
283                          " operation %d but we issued a request for %d\n",
284                          op, expected);
285                 return -EIO;
286         }
287         READ32(nfserr);
288         if (nfserr != NFS_OK)
289                 return -nfs_cb_stat_to_errno(nfserr);
290         return 0;
291 }
292
293 static int
294 nfs4_xdr_dec_cb_null(struct rpc_rqst *req, __be32 *p)
295 {
296         return 0;
297 }
298
299 static int
300 nfs4_xdr_dec_cb_recall(struct rpc_rqst *rqstp, __be32 *p)
301 {
302         struct xdr_stream xdr;
303         struct nfs4_cb_compound_hdr hdr;
304         int status;
305
306         xdr_init_decode(&xdr, &rqstp->rq_rcv_buf, p);
307         status = decode_cb_compound_hdr(&xdr, &hdr);
308         if (status)
309                 goto out;
310         status = decode_cb_op_hdr(&xdr, OP_CB_RECALL);
311 out:
312         return status;
313 }
314
315 /*
316  * RPC procedure tables
317  */
318 #define PROC(proc, call, argtype, restype)                              \
319 [NFSPROC4_CLNT_##proc] = {                                              \
320         .p_proc   = NFSPROC4_CB_##call,                                 \
321         .p_encode = (kxdrproc_t) nfs4_xdr_##argtype,                    \
322         .p_decode = (kxdrproc_t) nfs4_xdr_##restype,                    \
323         .p_arglen = NFS4_##argtype##_sz,                                \
324         .p_replen = NFS4_##restype##_sz,                                \
325         .p_statidx = NFSPROC4_CB_##call,                                \
326         .p_name   = #proc,                                              \
327 }
328
329 static struct rpc_procinfo     nfs4_cb_procedures[] = {
330     PROC(CB_NULL,      NULL,     enc_cb_null,     dec_cb_null),
331     PROC(CB_RECALL,    COMPOUND,   enc_cb_recall,      dec_cb_recall),
332 };
333
334 static struct rpc_version       nfs_cb_version4 = {
335         .number                 = 1,
336         .nrprocs                = ARRAY_SIZE(nfs4_cb_procedures),
337         .procs                  = nfs4_cb_procedures
338 };
339
340 static struct rpc_version *     nfs_cb_version[] = {
341         NULL,
342         &nfs_cb_version4,
343 };
344
345 static struct rpc_program cb_program;
346
347 static struct rpc_stat cb_stats = {
348                 .program        = &cb_program
349 };
350
351 #define NFS4_CALLBACK 0x40000000
352 static struct rpc_program cb_program = {
353                 .name           = "nfs4_cb",
354                 .number         = NFS4_CALLBACK,
355                 .nrvers         = ARRAY_SIZE(nfs_cb_version),
356                 .version        = nfs_cb_version,
357                 .stats          = &cb_stats,
358                 .pipe_dir_name  = "/nfsd4_cb",
359 };
360
361 /* Reference counting, callback cleanup, etc., all look racy as heck.
362  * And why is cb_set an atomic? */
363
364 static struct rpc_clnt *setup_callback_client(struct nfs4_client *clp)
365 {
366         struct sockaddr_in      addr;
367         struct nfs4_callback    *cb = &clp->cl_callback;
368         struct rpc_timeout      timeparms = {
369                 .to_initval     = (NFSD_LEASE_TIME/4) * HZ,
370                 .to_retries     = 5,
371                 .to_maxval      = (NFSD_LEASE_TIME/2) * HZ,
372                 .to_exponential = 1,
373         };
374         struct rpc_create_args args = {
375                 .protocol       = IPPROTO_TCP,
376                 .address        = (struct sockaddr *)&addr,
377                 .addrsize       = sizeof(addr),
378                 .timeout        = &timeparms,
379                 .program        = &cb_program,
380                 .prognumber     = cb->cb_prog,
381                 .version        = nfs_cb_version[1]->number,
382                 .authflavor     = clp->cl_flavor,
383                 .flags          = (RPC_CLNT_CREATE_NOPING | RPC_CLNT_CREATE_QUIET),
384                 .client_name    = clp->cl_principal,
385         };
386         struct rpc_clnt *client;
387
388         if (!clp->cl_principal && (clp->cl_flavor >= RPC_AUTH_GSS_KRB5))
389                 return ERR_PTR(-EINVAL);
390
391         /* Initialize address */
392         memset(&addr, 0, sizeof(addr));
393         addr.sin_family = AF_INET;
394         addr.sin_port = htons(cb->cb_port);
395         addr.sin_addr.s_addr = htonl(cb->cb_addr);
396
397         /* Create RPC client */
398         client = rpc_create(&args);
399         if (IS_ERR(client))
400                 dprintk("NFSD: couldn't create callback client: %ld\n",
401                         PTR_ERR(client));
402         return client;
403
404 }
405
406 static int do_probe_callback(void *data)
407 {
408         struct nfs4_client *clp = data;
409         struct nfs4_callback    *cb = &clp->cl_callback;
410         struct rpc_message msg = {
411                 .rpc_proc       = &nfs4_cb_procedures[NFSPROC4_CLNT_CB_NULL],
412                 .rpc_argp       = clp,
413         };
414         struct rpc_clnt *client;
415         int status;
416
417         client = setup_callback_client(clp);
418         if (IS_ERR(client)) {
419                 status = PTR_ERR(client);
420                 dprintk("NFSD: couldn't create callback client: %d\n",
421                                                                 status);
422                 goto out_err;
423         }
424
425         status = rpc_call_sync(client, &msg, RPC_TASK_SOFT);
426
427         if (status)
428                 goto out_release_client;
429
430         cb->cb_client = client;
431         atomic_set(&cb->cb_set, 1);
432         put_nfs4_client(clp);
433         return 0;
434 out_release_client:
435         rpc_shutdown_client(client);
436 out_err:
437         dprintk("NFSD: warning: no callback path to client %.*s: error %d\n",
438                 (int)clp->cl_name.len, clp->cl_name.data, status);
439         put_nfs4_client(clp);
440         return 0;
441 }
442
443 /*
444  * Set up the callback client and put a NFSPROC4_CB_NULL on the wire...
445  */
446 void
447 nfsd4_probe_callback(struct nfs4_client *clp)
448 {
449         struct task_struct *t;
450
451         BUG_ON(atomic_read(&clp->cl_callback.cb_set));
452
453         /* the task holds a reference to the nfs4_client struct */
454         atomic_inc(&clp->cl_count);
455
456         t = kthread_run(do_probe_callback, clp, "nfs4_cb_probe");
457
458         if (IS_ERR(t))
459                 atomic_dec(&clp->cl_count);
460
461         return;
462 }
463
464 /*
465  * called with dp->dl_count inc'ed.
466  */
467 void
468 nfsd4_cb_recall(struct nfs4_delegation *dp)
469 {
470         struct nfs4_client *clp = dp->dl_client;
471         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
472         struct nfs4_cb_recall *cbr = &dp->dl_recall;
473         struct rpc_message msg = {
474                 .rpc_proc = &nfs4_cb_procedures[NFSPROC4_CLNT_CB_RECALL],
475                 .rpc_argp = cbr,
476         };
477         int retries = 1;
478         int status = 0;
479
480         cbr->cbr_trunc = 0; /* XXX need to implement truncate optimization */
481         cbr->cbr_dp = dp;
482
483         status = rpc_call_sync(clnt, &msg, RPC_TASK_SOFT);
484         while (retries--) {
485                 switch (status) {
486                         case -EIO:
487                                 /* Network partition? */
488                                 atomic_set(&clp->cl_callback.cb_set, 0);
489                         case -EBADHANDLE:
490                         case -NFS4ERR_BAD_STATEID:
491                                 /* Race: client probably got cb_recall
492                                  * before open reply granting delegation */
493                                 break;
494                         default:
495                                 goto out_put_cred;
496                 }
497                 ssleep(2);
498                 status = rpc_call_sync(clnt, &msg, RPC_TASK_SOFT);
499         }
500 out_put_cred:
501         /*
502          * Success or failure, now we're either waiting for lease expiration
503          * or deleg_return.
504          */
505         put_nfs4_client(clp);
506         nfs4_put_delegation(dp);
507         return;
508 }