xref: /linux/fs/nfsd/nfs4callback.c (revision 24bce201d79807b668bf9d9e0aca801c5c0d5f78)
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 "trace.h"
42 #include "xdr4cb.h"
43 #include "xdr4.h"
44 
45 #define NFSDDBG_FACILITY                NFSDDBG_PROC
46 
47 static void nfsd4_mark_cb_fault(struct nfs4_client *, int reason);
48 
49 #define NFSPROC4_CB_NULL 0
50 #define NFSPROC4_CB_COMPOUND 1
51 
52 /* Index of predefined Linux callback client operations */
53 
54 struct nfs4_cb_compound_hdr {
55 	/* args */
56 	u32		ident;	/* minorversion 0 only */
57 	u32		nops;
58 	__be32		*nops_p;
59 	u32		minorversion;
60 	/* res */
61 	int		status;
62 };
63 
64 static __be32 *xdr_encode_empty_array(__be32 *p)
65 {
66 	*p++ = xdr_zero;
67 	return p;
68 }
69 
70 /*
71  * Encode/decode NFSv4 CB basic data types
72  *
73  * Basic NFSv4 callback data types are defined in section 15 of RFC
74  * 3530: "Network File System (NFS) version 4 Protocol" and section
75  * 20 of RFC 5661: "Network File System (NFS) Version 4 Minor Version
76  * 1 Protocol"
77  */
78 
79 /*
80  *	nfs_cb_opnum4
81  *
82  *	enum nfs_cb_opnum4 {
83  *		OP_CB_GETATTR		= 3,
84  *		  ...
85  *	};
86  */
87 enum nfs_cb_opnum4 {
88 	OP_CB_GETATTR			= 3,
89 	OP_CB_RECALL			= 4,
90 	OP_CB_LAYOUTRECALL		= 5,
91 	OP_CB_NOTIFY			= 6,
92 	OP_CB_PUSH_DELEG		= 7,
93 	OP_CB_RECALL_ANY		= 8,
94 	OP_CB_RECALLABLE_OBJ_AVAIL	= 9,
95 	OP_CB_RECALL_SLOT		= 10,
96 	OP_CB_SEQUENCE			= 11,
97 	OP_CB_WANTS_CANCELLED		= 12,
98 	OP_CB_NOTIFY_LOCK		= 13,
99 	OP_CB_NOTIFY_DEVICEID		= 14,
100 	OP_CB_OFFLOAD			= 15,
101 	OP_CB_ILLEGAL			= 10044
102 };
103 
104 static void encode_nfs_cb_opnum4(struct xdr_stream *xdr, enum nfs_cb_opnum4 op)
105 {
106 	__be32 *p;
107 
108 	p = xdr_reserve_space(xdr, 4);
109 	*p = cpu_to_be32(op);
110 }
111 
112 /*
113  * nfs_fh4
114  *
115  *	typedef opaque nfs_fh4<NFS4_FHSIZE>;
116  */
117 static void encode_nfs_fh4(struct xdr_stream *xdr, const struct knfsd_fh *fh)
118 {
119 	u32 length = fh->fh_size;
120 	__be32 *p;
121 
122 	BUG_ON(length > NFS4_FHSIZE);
123 	p = xdr_reserve_space(xdr, 4 + length);
124 	xdr_encode_opaque(p, &fh->fh_raw, length);
125 }
126 
127 /*
128  * stateid4
129  *
130  *	struct stateid4 {
131  *		uint32_t	seqid;
132  *		opaque		other[12];
133  *	};
134  */
135 static void encode_stateid4(struct xdr_stream *xdr, const stateid_t *sid)
136 {
137 	__be32 *p;
138 
139 	p = xdr_reserve_space(xdr, NFS4_STATEID_SIZE);
140 	*p++ = cpu_to_be32(sid->si_generation);
141 	xdr_encode_opaque_fixed(p, &sid->si_opaque, NFS4_STATEID_OTHER_SIZE);
142 }
143 
144 /*
145  * sessionid4
146  *
147  *	typedef opaque sessionid4[NFS4_SESSIONID_SIZE];
148  */
149 static void encode_sessionid4(struct xdr_stream *xdr,
150 			      const struct nfsd4_session *session)
151 {
152 	__be32 *p;
153 
154 	p = xdr_reserve_space(xdr, NFS4_MAX_SESSIONID_LEN);
155 	xdr_encode_opaque_fixed(p, session->se_sessionid.data,
156 					NFS4_MAX_SESSIONID_LEN);
157 }
158 
159 /*
160  * nfsstat4
161  */
162 static const struct {
163 	int stat;
164 	int errno;
165 } nfs_cb_errtbl[] = {
166 	{ NFS4_OK,		0		},
167 	{ NFS4ERR_PERM,		-EPERM		},
168 	{ NFS4ERR_NOENT,	-ENOENT		},
169 	{ NFS4ERR_IO,		-EIO		},
170 	{ NFS4ERR_NXIO,		-ENXIO		},
171 	{ NFS4ERR_ACCESS,	-EACCES		},
172 	{ NFS4ERR_EXIST,	-EEXIST		},
173 	{ NFS4ERR_XDEV,		-EXDEV		},
174 	{ NFS4ERR_NOTDIR,	-ENOTDIR	},
175 	{ NFS4ERR_ISDIR,	-EISDIR		},
176 	{ NFS4ERR_INVAL,	-EINVAL		},
177 	{ NFS4ERR_FBIG,		-EFBIG		},
178 	{ NFS4ERR_NOSPC,	-ENOSPC		},
179 	{ NFS4ERR_ROFS,		-EROFS		},
180 	{ NFS4ERR_MLINK,	-EMLINK		},
181 	{ NFS4ERR_NAMETOOLONG,	-ENAMETOOLONG	},
182 	{ NFS4ERR_NOTEMPTY,	-ENOTEMPTY	},
183 	{ NFS4ERR_DQUOT,	-EDQUOT		},
184 	{ NFS4ERR_STALE,	-ESTALE		},
185 	{ NFS4ERR_BADHANDLE,	-EBADHANDLE	},
186 	{ NFS4ERR_BAD_COOKIE,	-EBADCOOKIE	},
187 	{ NFS4ERR_NOTSUPP,	-ENOTSUPP	},
188 	{ NFS4ERR_TOOSMALL,	-ETOOSMALL	},
189 	{ NFS4ERR_SERVERFAULT,	-ESERVERFAULT	},
190 	{ NFS4ERR_BADTYPE,	-EBADTYPE	},
191 	{ NFS4ERR_LOCKED,	-EAGAIN		},
192 	{ NFS4ERR_RESOURCE,	-EREMOTEIO	},
193 	{ NFS4ERR_SYMLINK,	-ELOOP		},
194 	{ NFS4ERR_OP_ILLEGAL,	-EOPNOTSUPP	},
195 	{ NFS4ERR_DEADLOCK,	-EDEADLK	},
196 	{ -1,			-EIO		}
197 };
198 
199 /*
200  * If we cannot translate the error, the recovery routines should
201  * handle it.
202  *
203  * Note: remaining NFSv4 error codes have values > 10000, so should
204  * not conflict with native Linux error codes.
205  */
206 static int nfs_cb_stat_to_errno(int status)
207 {
208 	int i;
209 
210 	for (i = 0; nfs_cb_errtbl[i].stat != -1; i++) {
211 		if (nfs_cb_errtbl[i].stat == status)
212 			return nfs_cb_errtbl[i].errno;
213 	}
214 
215 	dprintk("NFSD: Unrecognized NFS CB status value: %u\n", status);
216 	return -status;
217 }
218 
219 static int decode_cb_op_status(struct xdr_stream *xdr,
220 			       enum nfs_cb_opnum4 expected, int *status)
221 {
222 	__be32 *p;
223 	u32 op;
224 
225 	p = xdr_inline_decode(xdr, 4 + 4);
226 	if (unlikely(p == NULL))
227 		goto out_overflow;
228 	op = be32_to_cpup(p++);
229 	if (unlikely(op != expected))
230 		goto out_unexpected;
231 	*status = nfs_cb_stat_to_errno(be32_to_cpup(p));
232 	return 0;
233 out_overflow:
234 	return -EIO;
235 out_unexpected:
236 	dprintk("NFSD: Callback server returned operation %d but "
237 		"we issued a request for %d\n", op, expected);
238 	return -EIO;
239 }
240 
241 /*
242  * CB_COMPOUND4args
243  *
244  *	struct CB_COMPOUND4args {
245  *		utf8str_cs	tag;
246  *		uint32_t	minorversion;
247  *		uint32_t	callback_ident;
248  *		nfs_cb_argop4	argarray<>;
249  *	};
250 */
251 static void encode_cb_compound4args(struct xdr_stream *xdr,
252 				    struct nfs4_cb_compound_hdr *hdr)
253 {
254 	__be32 * p;
255 
256 	p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4);
257 	p = xdr_encode_empty_array(p);		/* empty tag */
258 	*p++ = cpu_to_be32(hdr->minorversion);
259 	*p++ = cpu_to_be32(hdr->ident);
260 
261 	hdr->nops_p = p;
262 	*p = cpu_to_be32(hdr->nops);		/* argarray element count */
263 }
264 
265 /*
266  * Update argarray element count
267  */
268 static void encode_cb_nops(struct nfs4_cb_compound_hdr *hdr)
269 {
270 	BUG_ON(hdr->nops > NFS4_MAX_BACK_CHANNEL_OPS);
271 	*hdr->nops_p = cpu_to_be32(hdr->nops);
272 }
273 
274 /*
275  * CB_COMPOUND4res
276  *
277  *	struct CB_COMPOUND4res {
278  *		nfsstat4	status;
279  *		utf8str_cs	tag;
280  *		nfs_cb_resop4	resarray<>;
281  *	};
282  */
283 static int decode_cb_compound4res(struct xdr_stream *xdr,
284 				  struct nfs4_cb_compound_hdr *hdr)
285 {
286 	u32 length;
287 	__be32 *p;
288 
289 	p = xdr_inline_decode(xdr, 4 + 4);
290 	if (unlikely(p == NULL))
291 		goto out_overflow;
292 	hdr->status = be32_to_cpup(p++);
293 	/* Ignore the tag */
294 	length = be32_to_cpup(p++);
295 	p = xdr_inline_decode(xdr, length + 4);
296 	if (unlikely(p == NULL))
297 		goto out_overflow;
298 	p += XDR_QUADLEN(length);
299 	hdr->nops = be32_to_cpup(p);
300 	return 0;
301 out_overflow:
302 	return -EIO;
303 }
304 
305 /*
306  * CB_RECALL4args
307  *
308  *	struct CB_RECALL4args {
309  *		stateid4	stateid;
310  *		bool		truncate;
311  *		nfs_fh4		fh;
312  *	};
313  */
314 static void encode_cb_recall4args(struct xdr_stream *xdr,
315 				  const struct nfs4_delegation *dp,
316 				  struct nfs4_cb_compound_hdr *hdr)
317 {
318 	__be32 *p;
319 
320 	encode_nfs_cb_opnum4(xdr, OP_CB_RECALL);
321 	encode_stateid4(xdr, &dp->dl_stid.sc_stateid);
322 
323 	p = xdr_reserve_space(xdr, 4);
324 	*p++ = xdr_zero;			/* truncate */
325 
326 	encode_nfs_fh4(xdr, &dp->dl_stid.sc_file->fi_fhandle);
327 
328 	hdr->nops++;
329 }
330 
331 /*
332  * CB_SEQUENCE4args
333  *
334  *	struct CB_SEQUENCE4args {
335  *		sessionid4		csa_sessionid;
336  *		sequenceid4		csa_sequenceid;
337  *		slotid4			csa_slotid;
338  *		slotid4			csa_highest_slotid;
339  *		bool			csa_cachethis;
340  *		referring_call_list4	csa_referring_call_lists<>;
341  *	};
342  */
343 static void encode_cb_sequence4args(struct xdr_stream *xdr,
344 				    const struct nfsd4_callback *cb,
345 				    struct nfs4_cb_compound_hdr *hdr)
346 {
347 	struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
348 	__be32 *p;
349 
350 	if (hdr->minorversion == 0)
351 		return;
352 
353 	encode_nfs_cb_opnum4(xdr, OP_CB_SEQUENCE);
354 	encode_sessionid4(xdr, session);
355 
356 	p = xdr_reserve_space(xdr, 4 + 4 + 4 + 4 + 4);
357 	*p++ = cpu_to_be32(session->se_cb_seq_nr);	/* csa_sequenceid */
358 	*p++ = xdr_zero;			/* csa_slotid */
359 	*p++ = xdr_zero;			/* csa_highest_slotid */
360 	*p++ = xdr_zero;			/* csa_cachethis */
361 	xdr_encode_empty_array(p);		/* csa_referring_call_lists */
362 
363 	hdr->nops++;
364 }
365 
366 /*
367  * CB_SEQUENCE4resok
368  *
369  *	struct CB_SEQUENCE4resok {
370  *		sessionid4	csr_sessionid;
371  *		sequenceid4	csr_sequenceid;
372  *		slotid4		csr_slotid;
373  *		slotid4		csr_highest_slotid;
374  *		slotid4		csr_target_highest_slotid;
375  *	};
376  *
377  *	union CB_SEQUENCE4res switch (nfsstat4 csr_status) {
378  *	case NFS4_OK:
379  *		CB_SEQUENCE4resok	csr_resok4;
380  *	default:
381  *		void;
382  *	};
383  *
384  * Our current back channel implmentation supports a single backchannel
385  * with a single slot.
386  */
387 static int decode_cb_sequence4resok(struct xdr_stream *xdr,
388 				    struct nfsd4_callback *cb)
389 {
390 	struct nfsd4_session *session = cb->cb_clp->cl_cb_session;
391 	int status = -ESERVERFAULT;
392 	__be32 *p;
393 	u32 dummy;
394 
395 	/*
396 	 * If the server returns different values for sessionID, slotID or
397 	 * sequence number, the server is looney tunes.
398 	 */
399 	p = xdr_inline_decode(xdr, NFS4_MAX_SESSIONID_LEN + 4 + 4 + 4 + 4);
400 	if (unlikely(p == NULL))
401 		goto out_overflow;
402 
403 	if (memcmp(p, session->se_sessionid.data, NFS4_MAX_SESSIONID_LEN)) {
404 		dprintk("NFS: %s Invalid session id\n", __func__);
405 		goto out;
406 	}
407 	p += XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN);
408 
409 	dummy = be32_to_cpup(p++);
410 	if (dummy != session->se_cb_seq_nr) {
411 		dprintk("NFS: %s Invalid sequence number\n", __func__);
412 		goto out;
413 	}
414 
415 	dummy = be32_to_cpup(p++);
416 	if (dummy != 0) {
417 		dprintk("NFS: %s Invalid slotid\n", __func__);
418 		goto out;
419 	}
420 
421 	/*
422 	 * FIXME: process highest slotid and target highest slotid
423 	 */
424 	status = 0;
425 out:
426 	cb->cb_seq_status = status;
427 	return status;
428 out_overflow:
429 	status = -EIO;
430 	goto out;
431 }
432 
433 static int decode_cb_sequence4res(struct xdr_stream *xdr,
434 				  struct nfsd4_callback *cb)
435 {
436 	int status;
437 
438 	if (cb->cb_clp->cl_minorversion == 0)
439 		return 0;
440 
441 	status = decode_cb_op_status(xdr, OP_CB_SEQUENCE, &cb->cb_seq_status);
442 	if (unlikely(status || cb->cb_seq_status))
443 		return status;
444 
445 	return decode_cb_sequence4resok(xdr, cb);
446 }
447 
448 /*
449  * NFSv4.0 and NFSv4.1 XDR encode functions
450  *
451  * NFSv4.0 callback argument types are defined in section 15 of RFC
452  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
453  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
454  * Protocol".
455  */
456 
457 /*
458  * NB: Without this zero space reservation, callbacks over krb5p fail
459  */
460 static void nfs4_xdr_enc_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
461 				 const void *__unused)
462 {
463 	xdr_reserve_space(xdr, 0);
464 }
465 
466 /*
467  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
468  */
469 static void nfs4_xdr_enc_cb_recall(struct rpc_rqst *req, struct xdr_stream *xdr,
470 				   const void *data)
471 {
472 	const struct nfsd4_callback *cb = data;
473 	const struct nfs4_delegation *dp = cb_to_delegation(cb);
474 	struct nfs4_cb_compound_hdr hdr = {
475 		.ident = cb->cb_clp->cl_cb_ident,
476 		.minorversion = cb->cb_clp->cl_minorversion,
477 	};
478 
479 	encode_cb_compound4args(xdr, &hdr);
480 	encode_cb_sequence4args(xdr, cb, &hdr);
481 	encode_cb_recall4args(xdr, dp, &hdr);
482 	encode_cb_nops(&hdr);
483 }
484 
485 
486 /*
487  * NFSv4.0 and NFSv4.1 XDR decode functions
488  *
489  * NFSv4.0 callback result types are defined in section 15 of RFC
490  * 3530: "Network File System (NFS) version 4 Protocol" and section 20
491  * of RFC 5661:  "Network File System (NFS) Version 4 Minor Version 1
492  * Protocol".
493  */
494 
495 static int nfs4_xdr_dec_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
496 				void *__unused)
497 {
498 	return 0;
499 }
500 
501 /*
502  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
503  */
504 static int nfs4_xdr_dec_cb_recall(struct rpc_rqst *rqstp,
505 				  struct xdr_stream *xdr,
506 				  void *data)
507 {
508 	struct nfsd4_callback *cb = data;
509 	struct nfs4_cb_compound_hdr hdr;
510 	int status;
511 
512 	status = decode_cb_compound4res(xdr, &hdr);
513 	if (unlikely(status))
514 		return status;
515 
516 	status = decode_cb_sequence4res(xdr, cb);
517 	if (unlikely(status || cb->cb_seq_status))
518 		return status;
519 
520 	return decode_cb_op_status(xdr, OP_CB_RECALL, &cb->cb_status);
521 }
522 
523 #ifdef CONFIG_NFSD_PNFS
524 /*
525  * CB_LAYOUTRECALL4args
526  *
527  *	struct layoutrecall_file4 {
528  *		nfs_fh4         lor_fh;
529  *		offset4         lor_offset;
530  *		length4         lor_length;
531  *		stateid4        lor_stateid;
532  *	};
533  *
534  *	union layoutrecall4 switch(layoutrecall_type4 lor_recalltype) {
535  *	case LAYOUTRECALL4_FILE:
536  *		layoutrecall_file4 lor_layout;
537  *	case LAYOUTRECALL4_FSID:
538  *		fsid4              lor_fsid;
539  *	case LAYOUTRECALL4_ALL:
540  *		void;
541  *	};
542  *
543  *	struct CB_LAYOUTRECALL4args {
544  *		layouttype4             clora_type;
545  *		layoutiomode4           clora_iomode;
546  *		bool                    clora_changed;
547  *		layoutrecall4           clora_recall;
548  *	};
549  */
550 static void encode_cb_layout4args(struct xdr_stream *xdr,
551 				  const struct nfs4_layout_stateid *ls,
552 				  struct nfs4_cb_compound_hdr *hdr)
553 {
554 	__be32 *p;
555 
556 	BUG_ON(hdr->minorversion == 0);
557 
558 	p = xdr_reserve_space(xdr, 5 * 4);
559 	*p++ = cpu_to_be32(OP_CB_LAYOUTRECALL);
560 	*p++ = cpu_to_be32(ls->ls_layout_type);
561 	*p++ = cpu_to_be32(IOMODE_ANY);
562 	*p++ = cpu_to_be32(1);
563 	*p = cpu_to_be32(RETURN_FILE);
564 
565 	encode_nfs_fh4(xdr, &ls->ls_stid.sc_file->fi_fhandle);
566 
567 	p = xdr_reserve_space(xdr, 2 * 8);
568 	p = xdr_encode_hyper(p, 0);
569 	xdr_encode_hyper(p, NFS4_MAX_UINT64);
570 
571 	encode_stateid4(xdr, &ls->ls_recall_sid);
572 
573 	hdr->nops++;
574 }
575 
576 static void nfs4_xdr_enc_cb_layout(struct rpc_rqst *req,
577 				   struct xdr_stream *xdr,
578 				   const void *data)
579 {
580 	const struct nfsd4_callback *cb = data;
581 	const struct nfs4_layout_stateid *ls =
582 		container_of(cb, struct nfs4_layout_stateid, ls_recall);
583 	struct nfs4_cb_compound_hdr hdr = {
584 		.ident = 0,
585 		.minorversion = cb->cb_clp->cl_minorversion,
586 	};
587 
588 	encode_cb_compound4args(xdr, &hdr);
589 	encode_cb_sequence4args(xdr, cb, &hdr);
590 	encode_cb_layout4args(xdr, ls, &hdr);
591 	encode_cb_nops(&hdr);
592 }
593 
594 static int nfs4_xdr_dec_cb_layout(struct rpc_rqst *rqstp,
595 				  struct xdr_stream *xdr,
596 				  void *data)
597 {
598 	struct nfsd4_callback *cb = data;
599 	struct nfs4_cb_compound_hdr hdr;
600 	int status;
601 
602 	status = decode_cb_compound4res(xdr, &hdr);
603 	if (unlikely(status))
604 		return status;
605 
606 	status = decode_cb_sequence4res(xdr, cb);
607 	if (unlikely(status || cb->cb_seq_status))
608 		return status;
609 
610 	return decode_cb_op_status(xdr, OP_CB_LAYOUTRECALL, &cb->cb_status);
611 }
612 #endif /* CONFIG_NFSD_PNFS */
613 
614 static void encode_stateowner(struct xdr_stream *xdr, struct nfs4_stateowner *so)
615 {
616 	__be32	*p;
617 
618 	p = xdr_reserve_space(xdr, 8 + 4 + so->so_owner.len);
619 	p = xdr_encode_opaque_fixed(p, &so->so_client->cl_clientid, 8);
620 	xdr_encode_opaque(p, so->so_owner.data, so->so_owner.len);
621 }
622 
623 static void nfs4_xdr_enc_cb_notify_lock(struct rpc_rqst *req,
624 					struct xdr_stream *xdr,
625 					const void *data)
626 {
627 	const struct nfsd4_callback *cb = data;
628 	const struct nfsd4_blocked_lock *nbl =
629 		container_of(cb, struct nfsd4_blocked_lock, nbl_cb);
630 	struct nfs4_lockowner *lo = (struct nfs4_lockowner *)nbl->nbl_lock.fl_owner;
631 	struct nfs4_cb_compound_hdr hdr = {
632 		.ident = 0,
633 		.minorversion = cb->cb_clp->cl_minorversion,
634 	};
635 
636 	__be32 *p;
637 
638 	BUG_ON(hdr.minorversion == 0);
639 
640 	encode_cb_compound4args(xdr, &hdr);
641 	encode_cb_sequence4args(xdr, cb, &hdr);
642 
643 	p = xdr_reserve_space(xdr, 4);
644 	*p = cpu_to_be32(OP_CB_NOTIFY_LOCK);
645 	encode_nfs_fh4(xdr, &nbl->nbl_fh);
646 	encode_stateowner(xdr, &lo->lo_owner);
647 	hdr.nops++;
648 
649 	encode_cb_nops(&hdr);
650 }
651 
652 static int nfs4_xdr_dec_cb_notify_lock(struct rpc_rqst *rqstp,
653 					struct xdr_stream *xdr,
654 					void *data)
655 {
656 	struct nfsd4_callback *cb = data;
657 	struct nfs4_cb_compound_hdr hdr;
658 	int status;
659 
660 	status = decode_cb_compound4res(xdr, &hdr);
661 	if (unlikely(status))
662 		return status;
663 
664 	status = decode_cb_sequence4res(xdr, cb);
665 	if (unlikely(status || cb->cb_seq_status))
666 		return status;
667 
668 	return decode_cb_op_status(xdr, OP_CB_NOTIFY_LOCK, &cb->cb_status);
669 }
670 
671 /*
672  * struct write_response4 {
673  *	stateid4	wr_callback_id<1>;
674  *	length4		wr_count;
675  *	stable_how4	wr_committed;
676  *	verifier4	wr_writeverf;
677  * };
678  * union offload_info4 switch (nfsstat4 coa_status) {
679  *	case NFS4_OK:
680  *		write_response4	coa_resok4;
681  *	default:
682  *	length4		coa_bytes_copied;
683  * };
684  * struct CB_OFFLOAD4args {
685  *	nfs_fh4		coa_fh;
686  *	stateid4	coa_stateid;
687  *	offload_info4	coa_offload_info;
688  * };
689  */
690 static void encode_offload_info4(struct xdr_stream *xdr,
691 				 __be32 nfserr,
692 				 const struct nfsd4_copy *cp)
693 {
694 	__be32 *p;
695 
696 	p = xdr_reserve_space(xdr, 4);
697 	*p++ = nfserr;
698 	if (!nfserr) {
699 		p = xdr_reserve_space(xdr, 4 + 8 + 4 + NFS4_VERIFIER_SIZE);
700 		p = xdr_encode_empty_array(p);
701 		p = xdr_encode_hyper(p, cp->cp_res.wr_bytes_written);
702 		*p++ = cpu_to_be32(cp->cp_res.wr_stable_how);
703 		p = xdr_encode_opaque_fixed(p, cp->cp_res.wr_verifier.data,
704 					    NFS4_VERIFIER_SIZE);
705 	} else {
706 		p = xdr_reserve_space(xdr, 8);
707 		/* We always return success if bytes were written */
708 		p = xdr_encode_hyper(p, 0);
709 	}
710 }
711 
712 static void encode_cb_offload4args(struct xdr_stream *xdr,
713 				   __be32 nfserr,
714 				   const struct knfsd_fh *fh,
715 				   const struct nfsd4_copy *cp,
716 				   struct nfs4_cb_compound_hdr *hdr)
717 {
718 	__be32 *p;
719 
720 	p = xdr_reserve_space(xdr, 4);
721 	*p++ = cpu_to_be32(OP_CB_OFFLOAD);
722 	encode_nfs_fh4(xdr, fh);
723 	encode_stateid4(xdr, &cp->cp_res.cb_stateid);
724 	encode_offload_info4(xdr, nfserr, cp);
725 
726 	hdr->nops++;
727 }
728 
729 static void nfs4_xdr_enc_cb_offload(struct rpc_rqst *req,
730 				    struct xdr_stream *xdr,
731 				    const void *data)
732 {
733 	const struct nfsd4_callback *cb = data;
734 	const struct nfsd4_copy *cp =
735 		container_of(cb, struct nfsd4_copy, cp_cb);
736 	struct nfs4_cb_compound_hdr hdr = {
737 		.ident = 0,
738 		.minorversion = cb->cb_clp->cl_minorversion,
739 	};
740 
741 	encode_cb_compound4args(xdr, &hdr);
742 	encode_cb_sequence4args(xdr, cb, &hdr);
743 	encode_cb_offload4args(xdr, cp->nfserr, &cp->fh, cp, &hdr);
744 	encode_cb_nops(&hdr);
745 }
746 
747 static int nfs4_xdr_dec_cb_offload(struct rpc_rqst *rqstp,
748 				   struct xdr_stream *xdr,
749 				   void *data)
750 {
751 	struct nfsd4_callback *cb = data;
752 	struct nfs4_cb_compound_hdr hdr;
753 	int status;
754 
755 	status = decode_cb_compound4res(xdr, &hdr);
756 	if (unlikely(status))
757 		return status;
758 
759 	status = decode_cb_sequence4res(xdr, cb);
760 	if (unlikely(status || cb->cb_seq_status))
761 		return status;
762 
763 	return decode_cb_op_status(xdr, OP_CB_OFFLOAD, &cb->cb_status);
764 }
765 /*
766  * RPC procedure tables
767  */
768 #define PROC(proc, call, argtype, restype)				\
769 [NFSPROC4_CLNT_##proc] = {						\
770 	.p_proc    = NFSPROC4_CB_##call,				\
771 	.p_encode  = nfs4_xdr_enc_##argtype,		\
772 	.p_decode  = nfs4_xdr_dec_##restype,				\
773 	.p_arglen  = NFS4_enc_##argtype##_sz,				\
774 	.p_replen  = NFS4_dec_##restype##_sz,				\
775 	.p_statidx = NFSPROC4_CB_##call,				\
776 	.p_name    = #proc,						\
777 }
778 
779 static const struct rpc_procinfo nfs4_cb_procedures[] = {
780 	PROC(CB_NULL,	NULL,		cb_null,	cb_null),
781 	PROC(CB_RECALL,	COMPOUND,	cb_recall,	cb_recall),
782 #ifdef CONFIG_NFSD_PNFS
783 	PROC(CB_LAYOUT,	COMPOUND,	cb_layout,	cb_layout),
784 #endif
785 	PROC(CB_NOTIFY_LOCK,	COMPOUND,	cb_notify_lock,	cb_notify_lock),
786 	PROC(CB_OFFLOAD,	COMPOUND,	cb_offload,	cb_offload),
787 };
788 
789 static unsigned int nfs4_cb_counts[ARRAY_SIZE(nfs4_cb_procedures)];
790 static const struct rpc_version nfs_cb_version4 = {
791 /*
792  * Note on the callback rpc program version number: despite language in rfc
793  * 5661 section 18.36.3 requiring servers to use 4 in this field, the
794  * official xdr descriptions for both 4.0 and 4.1 specify version 1, and
795  * in practice that appears to be what implementations use.  The section
796  * 18.36.3 language is expected to be fixed in an erratum.
797  */
798 	.number			= 1,
799 	.nrprocs		= ARRAY_SIZE(nfs4_cb_procedures),
800 	.procs			= nfs4_cb_procedures,
801 	.counts			= nfs4_cb_counts,
802 };
803 
804 static const struct rpc_version *nfs_cb_version[2] = {
805 	[1] = &nfs_cb_version4,
806 };
807 
808 static const struct rpc_program cb_program;
809 
810 static struct rpc_stat cb_stats = {
811 	.program		= &cb_program
812 };
813 
814 #define NFS4_CALLBACK 0x40000000
815 static const struct rpc_program cb_program = {
816 	.name			= "nfs4_cb",
817 	.number			= NFS4_CALLBACK,
818 	.nrvers			= ARRAY_SIZE(nfs_cb_version),
819 	.version		= nfs_cb_version,
820 	.stats			= &cb_stats,
821 	.pipe_dir_name		= "nfsd4_cb",
822 };
823 
824 static int max_cb_time(struct net *net)
825 {
826 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
827 
828 	/*
829 	 * nfsd4_lease is set to at most one hour in __nfsd4_write_time,
830 	 * so we can use 32-bit math on it. Warn if that assumption
831 	 * ever stops being true.
832 	 */
833 	if (WARN_ON_ONCE(nn->nfsd4_lease > 3600))
834 		return 360 * HZ;
835 
836 	return max(((u32)nn->nfsd4_lease)/10, 1u) * HZ;
837 }
838 
839 static struct workqueue_struct *callback_wq;
840 
841 static bool nfsd4_queue_cb(struct nfsd4_callback *cb)
842 {
843 	return queue_work(callback_wq, &cb->cb_work);
844 }
845 
846 static void nfsd41_cb_inflight_begin(struct nfs4_client *clp)
847 {
848 	atomic_inc(&clp->cl_cb_inflight);
849 }
850 
851 static void nfsd41_cb_inflight_end(struct nfs4_client *clp)
852 {
853 
854 	if (atomic_dec_and_test(&clp->cl_cb_inflight))
855 		wake_up_var(&clp->cl_cb_inflight);
856 }
857 
858 static void nfsd41_cb_inflight_wait_complete(struct nfs4_client *clp)
859 {
860 	wait_var_event(&clp->cl_cb_inflight,
861 			!atomic_read(&clp->cl_cb_inflight));
862 }
863 
864 static const struct cred *get_backchannel_cred(struct nfs4_client *clp, struct rpc_clnt *client, struct nfsd4_session *ses)
865 {
866 	if (clp->cl_minorversion == 0) {
867 		client->cl_principal = clp->cl_cred.cr_targ_princ ?
868 			clp->cl_cred.cr_targ_princ : "nfs";
869 
870 		return get_cred(rpc_machine_cred());
871 	} else {
872 		struct cred *kcred;
873 
874 		kcred = prepare_kernel_cred(NULL);
875 		if (!kcred)
876 			return NULL;
877 
878 		kcred->uid = ses->se_cb_sec.uid;
879 		kcred->gid = ses->se_cb_sec.gid;
880 		return kcred;
881 	}
882 }
883 
884 static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn, struct nfsd4_session *ses)
885 {
886 	int maxtime = max_cb_time(clp->net);
887 	struct rpc_timeout	timeparms = {
888 		.to_initval	= maxtime,
889 		.to_retries	= 0,
890 		.to_maxval	= maxtime,
891 	};
892 	struct rpc_create_args args = {
893 		.net		= clp->net,
894 		.address	= (struct sockaddr *) &conn->cb_addr,
895 		.addrsize	= conn->cb_addrlen,
896 		.saddress	= (struct sockaddr *) &conn->cb_saddr,
897 		.timeout	= &timeparms,
898 		.program	= &cb_program,
899 		.version	= 1,
900 		.flags		= (RPC_CLNT_CREATE_NOPING | RPC_CLNT_CREATE_QUIET),
901 		.cred		= current_cred(),
902 	};
903 	struct rpc_clnt *client;
904 	const struct cred *cred;
905 
906 	if (clp->cl_minorversion == 0) {
907 		if (!clp->cl_cred.cr_principal &&
908 		    (clp->cl_cred.cr_flavor >= RPC_AUTH_GSS_KRB5)) {
909 			trace_nfsd_cb_setup_err(clp, -EINVAL);
910 			return -EINVAL;
911 		}
912 		args.client_name = clp->cl_cred.cr_principal;
913 		args.prognumber	= conn->cb_prog;
914 		args.protocol = XPRT_TRANSPORT_TCP;
915 		args.authflavor = clp->cl_cred.cr_flavor;
916 		clp->cl_cb_ident = conn->cb_ident;
917 	} else {
918 		if (!conn->cb_xprt)
919 			return -EINVAL;
920 		clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
921 		clp->cl_cb_session = ses;
922 		args.bc_xprt = conn->cb_xprt;
923 		args.prognumber = clp->cl_cb_session->se_cb_prog;
924 		args.protocol = conn->cb_xprt->xpt_class->xcl_ident |
925 				XPRT_TRANSPORT_BC;
926 		args.authflavor = ses->se_cb_sec.flavor;
927 	}
928 	/* Create RPC client */
929 	client = rpc_create(&args);
930 	if (IS_ERR(client)) {
931 		trace_nfsd_cb_setup_err(clp, PTR_ERR(client));
932 		return PTR_ERR(client);
933 	}
934 	cred = get_backchannel_cred(clp, client, ses);
935 	if (!cred) {
936 		trace_nfsd_cb_setup_err(clp, -ENOMEM);
937 		rpc_shutdown_client(client);
938 		return -ENOMEM;
939 	}
940 	clp->cl_cb_client = client;
941 	clp->cl_cb_cred = cred;
942 	rcu_read_lock();
943 	trace_nfsd_cb_setup(clp, rpc_peeraddr2str(client, RPC_DISPLAY_NETID),
944 			    args.authflavor);
945 	rcu_read_unlock();
946 	return 0;
947 }
948 
949 static void nfsd4_mark_cb_state(struct nfs4_client *clp, int newstate)
950 {
951 	if (clp->cl_cb_state != newstate) {
952 		clp->cl_cb_state = newstate;
953 		trace_nfsd_cb_state(clp);
954 	}
955 }
956 
957 static void nfsd4_mark_cb_down(struct nfs4_client *clp, int reason)
958 {
959 	if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
960 		return;
961 	nfsd4_mark_cb_state(clp, NFSD4_CB_DOWN);
962 }
963 
964 static void nfsd4_mark_cb_fault(struct nfs4_client *clp, int reason)
965 {
966 	if (test_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags))
967 		return;
968 	nfsd4_mark_cb_state(clp, NFSD4_CB_FAULT);
969 }
970 
971 static void nfsd4_cb_probe_done(struct rpc_task *task, void *calldata)
972 {
973 	struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
974 
975 	if (task->tk_status)
976 		nfsd4_mark_cb_down(clp, task->tk_status);
977 	else
978 		nfsd4_mark_cb_state(clp, NFSD4_CB_UP);
979 }
980 
981 static void nfsd4_cb_probe_release(void *calldata)
982 {
983 	struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
984 
985 	nfsd41_cb_inflight_end(clp);
986 
987 }
988 
989 static const struct rpc_call_ops nfsd4_cb_probe_ops = {
990 	/* XXX: release method to ensure we set the cb channel down if
991 	 * necessary on early failure? */
992 	.rpc_call_done = nfsd4_cb_probe_done,
993 	.rpc_release = nfsd4_cb_probe_release,
994 };
995 
996 /*
997  * Poke the callback thread to process any updates to the callback
998  * parameters, and send a null probe.
999  */
1000 void nfsd4_probe_callback(struct nfs4_client *clp)
1001 {
1002 	trace_nfsd_cb_probe(clp);
1003 	nfsd4_mark_cb_state(clp, NFSD4_CB_UNKNOWN);
1004 	set_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
1005 	nfsd4_run_cb(&clp->cl_cb_null);
1006 }
1007 
1008 void nfsd4_probe_callback_sync(struct nfs4_client *clp)
1009 {
1010 	nfsd4_probe_callback(clp);
1011 	flush_workqueue(callback_wq);
1012 }
1013 
1014 void nfsd4_change_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
1015 {
1016 	nfsd4_mark_cb_state(clp, NFSD4_CB_UNKNOWN);
1017 	spin_lock(&clp->cl_lock);
1018 	memcpy(&clp->cl_cb_conn, conn, sizeof(struct nfs4_cb_conn));
1019 	spin_unlock(&clp->cl_lock);
1020 }
1021 
1022 /*
1023  * There's currently a single callback channel slot.
1024  * If the slot is available, then mark it busy.  Otherwise, set the
1025  * thread for sleeping on the callback RPC wait queue.
1026  */
1027 static bool nfsd41_cb_get_slot(struct nfsd4_callback *cb, struct rpc_task *task)
1028 {
1029 	struct nfs4_client *clp = cb->cb_clp;
1030 
1031 	if (!cb->cb_holds_slot &&
1032 	    test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
1033 		rpc_sleep_on(&clp->cl_cb_waitq, task, NULL);
1034 		/* Race breaker */
1035 		if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
1036 			dprintk("%s slot is busy\n", __func__);
1037 			return false;
1038 		}
1039 		rpc_wake_up_queued_task(&clp->cl_cb_waitq, task);
1040 	}
1041 	cb->cb_holds_slot = true;
1042 	return true;
1043 }
1044 
1045 static void nfsd41_cb_release_slot(struct nfsd4_callback *cb)
1046 {
1047 	struct nfs4_client *clp = cb->cb_clp;
1048 
1049 	if (cb->cb_holds_slot) {
1050 		cb->cb_holds_slot = false;
1051 		clear_bit(0, &clp->cl_cb_slot_busy);
1052 		rpc_wake_up_next(&clp->cl_cb_waitq);
1053 	}
1054 }
1055 
1056 static void nfsd41_destroy_cb(struct nfsd4_callback *cb)
1057 {
1058 	struct nfs4_client *clp = cb->cb_clp;
1059 
1060 	nfsd41_cb_release_slot(cb);
1061 	if (cb->cb_ops && cb->cb_ops->release)
1062 		cb->cb_ops->release(cb);
1063 	nfsd41_cb_inflight_end(clp);
1064 }
1065 
1066 /*
1067  * TODO: cb_sequence should support referring call lists, cachethis, multiple
1068  * slots, and mark callback channel down on communication errors.
1069  */
1070 static void nfsd4_cb_prepare(struct rpc_task *task, void *calldata)
1071 {
1072 	struct nfsd4_callback *cb = calldata;
1073 	struct nfs4_client *clp = cb->cb_clp;
1074 	u32 minorversion = clp->cl_minorversion;
1075 
1076 	/*
1077 	 * cb_seq_status is only set in decode_cb_sequence4res,
1078 	 * and so will remain 1 if an rpc level failure occurs.
1079 	 */
1080 	cb->cb_seq_status = 1;
1081 	cb->cb_status = 0;
1082 	if (minorversion && !nfsd41_cb_get_slot(cb, task))
1083 		return;
1084 	rpc_call_start(task);
1085 }
1086 
1087 static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback *cb)
1088 {
1089 	struct nfs4_client *clp = cb->cb_clp;
1090 	struct nfsd4_session *session = clp->cl_cb_session;
1091 	bool ret = true;
1092 
1093 	if (!clp->cl_minorversion) {
1094 		/*
1095 		 * If the backchannel connection was shut down while this
1096 		 * task was queued, we need to resubmit it after setting up
1097 		 * a new backchannel connection.
1098 		 *
1099 		 * Note that if we lost our callback connection permanently
1100 		 * the submission code will error out, so we don't need to
1101 		 * handle that case here.
1102 		 */
1103 		if (RPC_SIGNALLED(task))
1104 			goto need_restart;
1105 
1106 		return true;
1107 	}
1108 
1109 	if (!cb->cb_holds_slot)
1110 		goto need_restart;
1111 
1112 	switch (cb->cb_seq_status) {
1113 	case 0:
1114 		/*
1115 		 * No need for lock, access serialized in nfsd4_cb_prepare
1116 		 *
1117 		 * RFC5661 20.9.3
1118 		 * If CB_SEQUENCE returns an error, then the state of the slot
1119 		 * (sequence ID, cached reply) MUST NOT change.
1120 		 */
1121 		++session->se_cb_seq_nr;
1122 		break;
1123 	case -ESERVERFAULT:
1124 		++session->se_cb_seq_nr;
1125 		fallthrough;
1126 	case 1:
1127 	case -NFS4ERR_BADSESSION:
1128 		nfsd4_mark_cb_fault(cb->cb_clp, cb->cb_seq_status);
1129 		ret = false;
1130 		break;
1131 	case -NFS4ERR_DELAY:
1132 		if (!rpc_restart_call(task))
1133 			goto out;
1134 
1135 		rpc_delay(task, 2 * HZ);
1136 		return false;
1137 	case -NFS4ERR_BADSLOT:
1138 		goto retry_nowait;
1139 	case -NFS4ERR_SEQ_MISORDERED:
1140 		if (session->se_cb_seq_nr != 1) {
1141 			session->se_cb_seq_nr = 1;
1142 			goto retry_nowait;
1143 		}
1144 		break;
1145 	default:
1146 		nfsd4_mark_cb_fault(cb->cb_clp, cb->cb_seq_status);
1147 		dprintk("%s: unprocessed error %d\n", __func__,
1148 			cb->cb_seq_status);
1149 	}
1150 
1151 	nfsd41_cb_release_slot(cb);
1152 	dprintk("%s: freed slot, new seqid=%d\n", __func__,
1153 		clp->cl_cb_session->se_cb_seq_nr);
1154 
1155 	if (RPC_SIGNALLED(task))
1156 		goto need_restart;
1157 out:
1158 	return ret;
1159 retry_nowait:
1160 	if (rpc_restart_call_prepare(task))
1161 		ret = false;
1162 	goto out;
1163 need_restart:
1164 	if (!test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags)) {
1165 		task->tk_status = 0;
1166 		cb->cb_need_restart = true;
1167 	}
1168 	return false;
1169 }
1170 
1171 static void nfsd4_cb_done(struct rpc_task *task, void *calldata)
1172 {
1173 	struct nfsd4_callback *cb = calldata;
1174 	struct nfs4_client *clp = cb->cb_clp;
1175 
1176 	if (!nfsd4_cb_sequence_done(task, cb))
1177 		return;
1178 
1179 	if (cb->cb_status) {
1180 		WARN_ON_ONCE(task->tk_status);
1181 		task->tk_status = cb->cb_status;
1182 	}
1183 
1184 	switch (cb->cb_ops->done(cb, task)) {
1185 	case 0:
1186 		task->tk_status = 0;
1187 		rpc_restart_call_prepare(task);
1188 		return;
1189 	case 1:
1190 		switch (task->tk_status) {
1191 		case -EIO:
1192 		case -ETIMEDOUT:
1193 		case -EACCES:
1194 			nfsd4_mark_cb_down(clp, task->tk_status);
1195 		}
1196 		break;
1197 	default:
1198 		BUG();
1199 	}
1200 }
1201 
1202 static void nfsd4_cb_release(void *calldata)
1203 {
1204 	struct nfsd4_callback *cb = calldata;
1205 
1206 	if (cb->cb_need_restart)
1207 		nfsd4_queue_cb(cb);
1208 	else
1209 		nfsd41_destroy_cb(cb);
1210 
1211 }
1212 
1213 static const struct rpc_call_ops nfsd4_cb_ops = {
1214 	.rpc_call_prepare = nfsd4_cb_prepare,
1215 	.rpc_call_done = nfsd4_cb_done,
1216 	.rpc_release = nfsd4_cb_release,
1217 };
1218 
1219 int nfsd4_create_callback_queue(void)
1220 {
1221 	callback_wq = alloc_ordered_workqueue("nfsd4_callbacks", 0);
1222 	if (!callback_wq)
1223 		return -ENOMEM;
1224 	return 0;
1225 }
1226 
1227 void nfsd4_destroy_callback_queue(void)
1228 {
1229 	destroy_workqueue(callback_wq);
1230 }
1231 
1232 /* must be called under the state lock */
1233 void nfsd4_shutdown_callback(struct nfs4_client *clp)
1234 {
1235 	if (clp->cl_cb_state != NFSD4_CB_UNKNOWN)
1236 		trace_nfsd_cb_shutdown(clp);
1237 
1238 	set_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags);
1239 	/*
1240 	 * Note this won't actually result in a null callback;
1241 	 * instead, nfsd4_run_cb_null() will detect the killed
1242 	 * client, destroy the rpc client, and stop:
1243 	 */
1244 	nfsd4_run_cb(&clp->cl_cb_null);
1245 	flush_workqueue(callback_wq);
1246 	nfsd41_cb_inflight_wait_complete(clp);
1247 }
1248 
1249 /* requires cl_lock: */
1250 static struct nfsd4_conn * __nfsd4_find_backchannel(struct nfs4_client *clp)
1251 {
1252 	struct nfsd4_session *s;
1253 	struct nfsd4_conn *c;
1254 
1255 	list_for_each_entry(s, &clp->cl_sessions, se_perclnt) {
1256 		list_for_each_entry(c, &s->se_conns, cn_persession) {
1257 			if (c->cn_flags & NFS4_CDFC4_BACK)
1258 				return c;
1259 		}
1260 	}
1261 	return NULL;
1262 }
1263 
1264 /*
1265  * Note there isn't a lot of locking in this code; instead we depend on
1266  * the fact that it is run from the callback_wq, which won't run two
1267  * work items at once.  So, for example, callback_wq handles all access
1268  * of cl_cb_client and all calls to rpc_create or rpc_shutdown_client.
1269  */
1270 static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
1271 {
1272 	struct nfs4_cb_conn conn;
1273 	struct nfs4_client *clp = cb->cb_clp;
1274 	struct nfsd4_session *ses = NULL;
1275 	struct nfsd4_conn *c;
1276 	int err;
1277 
1278 	/*
1279 	 * This is either an update, or the client dying; in either case,
1280 	 * kill the old client:
1281 	 */
1282 	if (clp->cl_cb_client) {
1283 		rpc_shutdown_client(clp->cl_cb_client);
1284 		clp->cl_cb_client = NULL;
1285 		put_cred(clp->cl_cb_cred);
1286 		clp->cl_cb_cred = NULL;
1287 	}
1288 	if (clp->cl_cb_conn.cb_xprt) {
1289 		svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1290 		clp->cl_cb_conn.cb_xprt = NULL;
1291 	}
1292 	if (test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags))
1293 		return;
1294 	spin_lock(&clp->cl_lock);
1295 	/*
1296 	 * Only serialized callback code is allowed to clear these
1297 	 * flags; main nfsd code can only set them:
1298 	 */
1299 	BUG_ON(!(clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK));
1300 	clear_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
1301 	memcpy(&conn, &cb->cb_clp->cl_cb_conn, sizeof(struct nfs4_cb_conn));
1302 	c = __nfsd4_find_backchannel(clp);
1303 	if (c) {
1304 		svc_xprt_get(c->cn_xprt);
1305 		conn.cb_xprt = c->cn_xprt;
1306 		ses = c->cn_session;
1307 	}
1308 	spin_unlock(&clp->cl_lock);
1309 
1310 	err = setup_callback_client(clp, &conn, ses);
1311 	if (err) {
1312 		nfsd4_mark_cb_down(clp, err);
1313 		if (c)
1314 			svc_xprt_put(c->cn_xprt);
1315 		return;
1316 	}
1317 }
1318 
1319 static void
1320 nfsd4_run_cb_work(struct work_struct *work)
1321 {
1322 	struct nfsd4_callback *cb =
1323 		container_of(work, struct nfsd4_callback, cb_work);
1324 	struct nfs4_client *clp = cb->cb_clp;
1325 	struct rpc_clnt *clnt;
1326 	int flags;
1327 
1328 	if (cb->cb_need_restart) {
1329 		cb->cb_need_restart = false;
1330 	} else {
1331 		if (cb->cb_ops && cb->cb_ops->prepare)
1332 			cb->cb_ops->prepare(cb);
1333 	}
1334 
1335 	if (clp->cl_flags & NFSD4_CLIENT_CB_FLAG_MASK)
1336 		nfsd4_process_cb_update(cb);
1337 
1338 	clnt = clp->cl_cb_client;
1339 	if (!clnt) {
1340 		/* Callback channel broken, or client killed; give up: */
1341 		nfsd41_destroy_cb(cb);
1342 		return;
1343 	}
1344 
1345 	/*
1346 	 * Don't send probe messages for 4.1 or later.
1347 	 */
1348 	if (!cb->cb_ops && clp->cl_minorversion) {
1349 		nfsd4_mark_cb_state(clp, NFSD4_CB_UP);
1350 		nfsd41_destroy_cb(cb);
1351 		return;
1352 	}
1353 
1354 	cb->cb_msg.rpc_cred = clp->cl_cb_cred;
1355 	flags = clp->cl_minorversion ? RPC_TASK_NOCONNECT : RPC_TASK_SOFTCONN;
1356 	rpc_call_async(clnt, &cb->cb_msg, RPC_TASK_SOFT | flags,
1357 			cb->cb_ops ? &nfsd4_cb_ops : &nfsd4_cb_probe_ops, cb);
1358 }
1359 
1360 void nfsd4_init_cb(struct nfsd4_callback *cb, struct nfs4_client *clp,
1361 		const struct nfsd4_callback_ops *ops, enum nfsd4_cb_op op)
1362 {
1363 	cb->cb_clp = clp;
1364 	cb->cb_msg.rpc_proc = &nfs4_cb_procedures[op];
1365 	cb->cb_msg.rpc_argp = cb;
1366 	cb->cb_msg.rpc_resp = cb;
1367 	cb->cb_ops = ops;
1368 	INIT_WORK(&cb->cb_work, nfsd4_run_cb_work);
1369 	cb->cb_seq_status = 1;
1370 	cb->cb_status = 0;
1371 	cb->cb_need_restart = false;
1372 	cb->cb_holds_slot = false;
1373 }
1374 
1375 void nfsd4_run_cb(struct nfsd4_callback *cb)
1376 {
1377 	struct nfs4_client *clp = cb->cb_clp;
1378 
1379 	nfsd41_cb_inflight_begin(clp);
1380 	if (!nfsd4_queue_cb(cb))
1381 		nfsd41_cb_inflight_end(clp);
1382 }
1383