mon.c 14.2 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10
/*
 * linux/fs/lockd/mon.c
 *
 * The kernel statd client.
 *
 * Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de>
 */

#include <linux/types.h>
#include <linux/kernel.h>
11
#include <linux/ktime.h>
12
#include <linux/slab.h>
13

L
Linus Torvalds 已提交
14
#include <linux/sunrpc/clnt.h>
15
#include <linux/sunrpc/addr.h>
16
#include <linux/sunrpc/xprtsock.h>
L
Linus Torvalds 已提交
17 18 19
#include <linux/sunrpc/svc.h>
#include <linux/lockd/lockd.h>

20 21
#include <asm/unaligned.h>

22 23
#include "netns.h"

L
Linus Torvalds 已提交
24
#define NLMDBG_FACILITY		NLMDBG_MONITOR
25 26 27 28 29 30 31 32 33 34 35 36
#define NSM_PROGRAM		100024
#define NSM_VERSION		1

enum {
	NSMPROC_NULL,
	NSMPROC_STAT,
	NSMPROC_MON,
	NSMPROC_UNMON,
	NSMPROC_UNMON_ALL,
	NSMPROC_SIMU_CRASH,
	NSMPROC_NOTIFY,
};
L
Linus Torvalds 已提交
37

38
struct nsm_args {
39
	struct nsm_private	*priv;
40 41 42 43 44
	u32			prog;		/* RPC callback info */
	u32			vers;
	u32			proc;

	char			*mon_name;
45
	const char		*nodename;
46 47 48 49 50 51 52
};

struct nsm_res {
	u32			status;
	u32			state;
};

53
static const struct rpc_program	nsm_program;
54
static				DEFINE_SPINLOCK(nsm_lock);
L
Linus Torvalds 已提交
55 56 57 58

/*
 * Local NSM state
 */
59
u32	__read_mostly		nsm_local_state;
60
bool	__read_mostly		nsm_use_hostnames;
L
Linus Torvalds 已提交
61

62 63 64 65 66
static inline struct sockaddr *nsm_addr(const struct nsm_handle *nsm)
{
	return (struct sockaddr *)&nsm->sm_addr;
}

67
static struct rpc_clnt *nsm_create(struct net *net, const char *nodename)
C
Chuck Lever 已提交
68 69 70 71 72 73
{
	struct sockaddr_in sin = {
		.sin_family		= AF_INET,
		.sin_addr.s_addr	= htonl(INADDR_LOOPBACK),
	};
	struct rpc_create_args args = {
74
		.net			= net,
75
		.protocol		= XPRT_TRANSPORT_TCP,
C
Chuck Lever 已提交
76 77 78
		.address		= (struct sockaddr *)&sin,
		.addrsize		= sizeof(sin),
		.servername		= "rpc.statd",
79
		.nodename		= nodename,
C
Chuck Lever 已提交
80 81 82
		.program		= &nsm_program,
		.version		= NSM_VERSION,
		.authflavor		= RPC_AUTH_NULL,
83
		.flags			= RPC_CLNT_CREATE_NOPING,
C
Chuck Lever 已提交
84 85 86 87 88
	};

	return rpc_create(&args);
}

89
static int nsm_mon_unmon(struct nsm_handle *nsm, u32 proc, struct nsm_res *res,
90
			 const struct nlm_host *host)
L
Linus Torvalds 已提交
91 92
{
	int		status;
93
	struct rpc_clnt *clnt;
94
	struct nsm_args args = {
95
		.priv		= &nsm->sm_priv,
96 97 98
		.prog		= NLM_PROGRAM,
		.vers		= 3,
		.proc		= NLMPROC_NSM_NOTIFY,
99
		.mon_name	= nsm->sm_mon_name,
100
		.nodename	= host->nodename,
101
	};
C
Chuck Lever 已提交
102 103 104 105
	struct rpc_message msg = {
		.rpc_argp	= &args,
		.rpc_resp	= res,
	};
L
Linus Torvalds 已提交
106 107 108

	memset(res, 0, sizeof(*res));

109 110 111 112 113 114 115
	clnt = nsm_create(host->net, host->nodename);
	if (IS_ERR(clnt)) {
		dprintk("lockd: failed to create NSM upcall transport, "
			"status=%ld, net=%p\n", PTR_ERR(clnt), host->net);
		return PTR_ERR(clnt);
	}

C
Chuck Lever 已提交
116
	msg.rpc_proc = &clnt->cl_procinfo[proc];
117
	status = rpc_call_sync(clnt, &msg, RPC_TASK_SOFTCONN);
118 119 120 121 122 123
	if (status == -ECONNREFUSED) {
		dprintk("lockd:	NSM upcall RPC failed, status=%d, forcing rebind\n",
				status);
		rpc_force_rebind(clnt);
		status = rpc_call_sync(clnt, &msg, RPC_TASK_SOFTCONN);
	}
L
Linus Torvalds 已提交
124
	if (status < 0)
125 126
		dprintk("lockd: NSM upcall RPC failed, status=%d\n",
				status);
L
Linus Torvalds 已提交
127 128
	else
		status = 0;
129 130

	rpc_shutdown_client(clnt);
L
Linus Torvalds 已提交
131 132 133
	return status;
}

134 135 136 137 138 139 140 141 142 143
/**
 * nsm_monitor - Notify a peer in case we reboot
 * @host: pointer to nlm_host of peer to notify
 *
 * If this peer is not already monitored, this function sends an
 * upcall to the local rpc.statd to record the name/address of
 * the peer to notify in case we reboot.
 *
 * Returns zero if the peer is monitored by the local rpc.statd;
 * otherwise a negative errno value is returned.
L
Linus Torvalds 已提交
144
 */
145
int nsm_monitor(const struct nlm_host *host)
L
Linus Torvalds 已提交
146
{
147
	struct nsm_handle *nsm = host->h_nsmhandle;
L
Linus Torvalds 已提交
148 149 150
	struct nsm_res	res;
	int		status;

151
	dprintk("lockd: nsm_monitor(%s)\n", nsm->sm_name);
152 153

	if (nsm->sm_monitored)
154
		return 0;
L
Linus Torvalds 已提交
155

156 157 158 159 160 161
	/*
	 * Choose whether to record the caller_name or IP address of
	 * this peer in the local rpc.statd's database.
	 */
	nsm->sm_mon_name = nsm_use_hostnames ? nsm->sm_name : nsm->sm_addrbuf;

162
	status = nsm_mon_unmon(nsm, NSMPROC_MON, &res, host);
163
	if (unlikely(res.status != 0))
164
		status = -EIO;
165
	if (unlikely(status < 0)) {
166
		pr_notice_ratelimited("lockd: cannot monitor %s\n", nsm->sm_name);
167 168 169 170 171 172 173 174 175
		return status;
	}

	nsm->sm_monitored = 1;
	if (unlikely(nsm_local_state != res.state)) {
		nsm_local_state = res.state;
		dprintk("lockd: NSM state changed to %d\n", nsm_local_state);
	}
	return 0;
L
Linus Torvalds 已提交
176 177
}

178 179 180 181 182 183 184
/**
 * nsm_unmonitor - Unregister peer notification
 * @host: pointer to nlm_host of peer to stop monitoring
 *
 * If this peer is monitored, this function sends an upcall to
 * tell the local rpc.statd not to send this peer a notification
 * when we reboot.
L
Linus Torvalds 已提交
185
 */
186
void nsm_unmonitor(const struct nlm_host *host)
L
Linus Torvalds 已提交
187
{
188
	struct nsm_handle *nsm = host->h_nsmhandle;
L
Linus Torvalds 已提交
189
	struct nsm_res	res;
190
	int status;
L
Linus Torvalds 已提交
191

192 193
	if (atomic_read(&nsm->sm_count) == 1
	 && nsm->sm_monitored && !nsm->sm_sticky) {
194
		dprintk("lockd: nsm_unmonitor(%s)\n", nsm->sm_name);
195

196
		status = nsm_mon_unmon(nsm, NSMPROC_UNMON, &res, host);
197 198
		if (res.status != 0)
			status = -EIO;
199
		if (status < 0)
200
			printk(KERN_NOTICE "lockd: cannot unmonitor %s\n",
201
					nsm->sm_name);
202 203
		else
			nsm->sm_monitored = 0;
204
	}
L
Linus Torvalds 已提交
205 206
}

207 208
static struct nsm_handle *nsm_lookup_hostname(const struct list_head *nsm_handles,
					const char *hostname, const size_t len)
C
Chuck Lever 已提交
209 210 211
{
	struct nsm_handle *nsm;

212
	list_for_each_entry(nsm, nsm_handles, sm_link)
C
Chuck Lever 已提交
213 214 215 216 217 218
		if (strlen(nsm->sm_name) == len &&
		    memcmp(nsm->sm_name, hostname, len) == 0)
			return nsm;
	return NULL;
}

219 220
static struct nsm_handle *nsm_lookup_addr(const struct list_head *nsm_handles,
					const struct sockaddr *sap)
221 222 223
{
	struct nsm_handle *nsm;

224
	list_for_each_entry(nsm, nsm_handles, sm_link)
225
		if (rpc_cmp_addr(nsm_addr(nsm), sap))
226 227 228 229
			return nsm;
	return NULL;
}

230 231
static struct nsm_handle *nsm_lookup_priv(const struct list_head *nsm_handles,
					const struct nsm_private *priv)
C
Chuck Lever 已提交
232 233 234
{
	struct nsm_handle *nsm;

235
	list_for_each_entry(nsm, nsm_handles, sm_link)
C
Chuck Lever 已提交
236 237 238 239 240 241
		if (memcmp(nsm->sm_priv.data, priv->data,
					sizeof(priv->data)) == 0)
			return nsm;
	return NULL;
}

242 243 244 245 246 247
/*
 * Construct a unique cookie to match this nsm_handle to this monitored
 * host.  It is passed to the local rpc.statd via NSMPROC_MON, and
 * returned via NLMPROC_SM_NOTIFY, in the "priv" field of these
 * requests.
 *
248 249 250 251 252 253 254 255 256 257
 * The NSM protocol requires that these cookies be unique while the
 * system is running.  We prefer a stronger requirement of making them
 * unique across reboots.  If user space bugs cause a stale cookie to
 * be sent to the kernel, it could cause the wrong host to lose its
 * lock state if cookies were not unique across reboots.
 *
 * The cookies are exposed only to local user space via loopback.  They
 * do not appear on the physical network.  If we want greater security
 * for some reason, nsm_init_private() could perform a one-way hash to
 * obscure the contents of the cookie.
258 259 260
 */
static void nsm_init_private(struct nsm_handle *nsm)
{
261
	u64 *p = (u64 *)&nsm->sm_priv.data;
262
	s64 ns;
263

T
Thomas Gleixner 已提交
264
	ns = ktime_get_ns();
265 266
	put_unaligned(ns, p);
	put_unaligned((unsigned long)nsm, p + 1);
267 268
}

269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284
static struct nsm_handle *nsm_create_handle(const struct sockaddr *sap,
					    const size_t salen,
					    const char *hostname,
					    const size_t hostname_len)
{
	struct nsm_handle *new;

	new = kzalloc(sizeof(*new) + hostname_len + 1, GFP_KERNEL);
	if (unlikely(new == NULL))
		return NULL;

	atomic_set(&new->sm_count, 1);
	new->sm_name = (char *)(new + 1);
	memcpy(nsm_addr(new), sap, salen);
	new->sm_addrlen = salen;
	nsm_init_private(new);
285 286 287 288 289

	if (rpc_ntop(nsm_addr(new), new->sm_addrbuf,
					sizeof(new->sm_addrbuf)) == 0)
		(void)snprintf(new->sm_addrbuf, sizeof(new->sm_addrbuf),
				"unsupported address family");
290 291 292 293 294 295
	memcpy(new->sm_name, hostname, hostname_len);
	new->sm_name[hostname_len] = '\0';

	return new;
}

296
/**
297
 * nsm_get_handle - Find or create a cached nsm_handle
298
 * @net: network namespace
299 300 301 302 303
 * @sap: pointer to socket address of handle to find
 * @salen: length of socket address
 * @hostname: pointer to C string containing hostname to find
 * @hostname_len: length of C string
 *
304
 * Behavior is modulated by the global nsm_use_hostnames variable.
305
 *
306 307 308 309
 * Returns a cached nsm_handle after bumping its ref count, or
 * returns a fresh nsm_handle if a handle that matches @sap and/or
 * @hostname cannot be found in the handle cache.  Returns NULL if
 * an error occurs.
310
 */
311 312
struct nsm_handle *nsm_get_handle(const struct net *net,
				  const struct sockaddr *sap,
313 314
				  const size_t salen, const char *hostname,
				  const size_t hostname_len)
315
{
316
	struct nsm_handle *cached, *new = NULL;
317
	struct lockd_net *ln = net_generic(net, lockd_net_id);
318 319 320 321 322 323 324 325 326 327 328 329

	if (hostname && memchr(hostname, '/', hostname_len) != NULL) {
		if (printk_ratelimit()) {
			printk(KERN_WARNING "Invalid hostname \"%.*s\" "
					    "in NFS lock request\n",
				(int)hostname_len, hostname);
		}
		return NULL;
	}

retry:
	spin_lock(&nsm_lock);
330 331

	if (nsm_use_hostnames && hostname != NULL)
332 333
		cached = nsm_lookup_hostname(&ln->nsm_handles,
					hostname, hostname_len);
334
	else
335
		cached = nsm_lookup_addr(&ln->nsm_handles, sap);
336 337 338 339 340 341 342 343 344 345

	if (cached != NULL) {
		atomic_inc(&cached->sm_count);
		spin_unlock(&nsm_lock);
		kfree(new);
		dprintk("lockd: found nsm_handle for %s (%s), "
				"cnt %d\n", cached->sm_name,
				cached->sm_addrbuf,
				atomic_read(&cached->sm_count));
		return cached;
346
	}
347 348

	if (new != NULL) {
349
		list_add(&new->sm_link, &ln->nsm_handles);
350
		spin_unlock(&nsm_lock);
351
		dprintk("lockd: created nsm_handle for %s (%s)\n",
352 353
				new->sm_name, new->sm_addrbuf);
		return new;
354
	}
355

356 357
	spin_unlock(&nsm_lock);

358 359
	new = nsm_create_handle(sap, salen, hostname, hostname_len);
	if (unlikely(new == NULL))
360 361 362 363
		return NULL;
	goto retry;
}

C
Chuck Lever 已提交
364 365
/**
 * nsm_reboot_lookup - match NLMPROC_SM_NOTIFY arguments to an nsm_handle
366
 * @net:  network namespace
C
Chuck Lever 已提交
367 368
 * @info: pointer to NLMPROC_SM_NOTIFY arguments
 *
369 370 371
 * Returns a matching nsm_handle if found in the nsm cache. The returned
 * nsm_handle's reference count is bumped. Otherwise returns NULL if some
 * error occurred.
C
Chuck Lever 已提交
372
 */
373 374
struct nsm_handle *nsm_reboot_lookup(const struct net *net,
				const struct nlm_reboot *info)
C
Chuck Lever 已提交
375 376
{
	struct nsm_handle *cached;
377
	struct lockd_net *ln = net_generic(net, lockd_net_id);
C
Chuck Lever 已提交
378 379 380

	spin_lock(&nsm_lock);

381
	cached = nsm_lookup_priv(&ln->nsm_handles, &info->priv);
C
Chuck Lever 已提交
382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
	if (unlikely(cached == NULL)) {
		spin_unlock(&nsm_lock);
		dprintk("lockd: never saw rebooted peer '%.*s' before\n",
				info->len, info->mon);
		return cached;
	}

	atomic_inc(&cached->sm_count);
	spin_unlock(&nsm_lock);

	dprintk("lockd: host %s (%s) rebooted, cnt %d\n",
			cached->sm_name, cached->sm_addrbuf,
			atomic_read(&cached->sm_count));
	return cached;
}

398 399 400 401 402 403 404 405 406 407
/**
 * nsm_release - Release an NSM handle
 * @nsm: pointer to handle to be released
 *
 */
void nsm_release(struct nsm_handle *nsm)
{
	if (atomic_dec_and_lock(&nsm->sm_count, &nsm_lock)) {
		list_del(&nsm->sm_link);
		spin_unlock(&nsm_lock);
408 409
		dprintk("lockd: destroyed nsm_handle for %s (%s)\n",
				nsm->sm_name, nsm->sm_addrbuf);
410 411 412 413
		kfree(nsm);
	}
}

L
Linus Torvalds 已提交
414 415
/*
 * XDR functions for NSM.
416 417 418
 *
 * See http://www.opengroup.org/ for details on the Network
 * Status Monitor wire protocol.
L
Linus Torvalds 已提交
419 420
 */

421
static void encode_nsm_string(struct xdr_stream *xdr, const char *string)
422
{
423 424 425
	const u32 len = strlen(string);
	__be32 *p;

426
	p = xdr_reserve_space(xdr, 4 + len);
427
	xdr_encode_opaque(p, string, len);
428 429
}

430 431 432
/*
 * "mon_name" specifies the host to be monitored.
 */
433
static void encode_mon_name(struct xdr_stream *xdr, const struct nsm_args *argp)
434
{
435
	encode_nsm_string(xdr, argp->mon_name);
436 437
}

438 439 440
/*
 * The "my_id" argument specifies the hostname and RPC procedure
 * to be called when the status manager receives notification
441
 * (via the NLMPROC_SM_NOTIFY call) that the state of host "mon_name"
442 443
 * has changed.
 */
444
static void encode_my_id(struct xdr_stream *xdr, const struct nsm_args *argp)
445
{
446 447
	__be32 *p;

448
	encode_nsm_string(xdr, argp->nodename);
449 450 451 452
	p = xdr_reserve_space(xdr, 4 + 4 + 4);
	*p++ = cpu_to_be32(argp->prog);
	*p++ = cpu_to_be32(argp->vers);
	*p = cpu_to_be32(argp->proc);
453 454
}

455 456
/*
 * The "mon_id" argument specifies the non-private arguments
457
 * of an NSMPROC_MON or NSMPROC_UNMON call.
458
 */
459
static void encode_mon_id(struct xdr_stream *xdr, const struct nsm_args *argp)
460
{
461 462
	encode_mon_name(xdr, argp);
	encode_my_id(xdr, argp);
463 464
}

465 466
/*
 * The "priv" argument may contain private information required
467 468
 * by the NSMPROC_MON call. This information will be supplied in the
 * NLMPROC_SM_NOTIFY call.
469
 */
470
static void encode_priv(struct xdr_stream *xdr, const struct nsm_args *argp)
471
{
472 473 474
	__be32 *p;

	p = xdr_reserve_space(xdr, SM_PRIV_SIZE);
475
	xdr_encode_opaque_fixed(p, argp->priv->data, SM_PRIV_SIZE);
476 477
}

478 479
static void nsm_xdr_enc_mon(struct rpc_rqst *req, struct xdr_stream *xdr,
			    const struct nsm_args *argp)
L
Linus Torvalds 已提交
480
{
481 482
	encode_mon_id(xdr, argp);
	encode_priv(xdr, argp);
L
Linus Torvalds 已提交
483 484
}

485 486
static void nsm_xdr_enc_unmon(struct rpc_rqst *req, struct xdr_stream *xdr,
			      const struct nsm_args *argp)
L
Linus Torvalds 已提交
487
{
488
	encode_mon_id(xdr, argp);
L
Linus Torvalds 已提交
489 490
}

491 492 493
static int nsm_xdr_dec_stat_res(struct rpc_rqst *rqstp,
				struct xdr_stream *xdr,
				struct nsm_res *resp)
L
Linus Torvalds 已提交
494
{
495
	__be32 *p;
496

497
	p = xdr_inline_decode(xdr, 4 + 4);
498 499
	if (unlikely(p == NULL))
		return -EIO;
500 501
	resp->status = be32_to_cpup(p++);
	resp->state = be32_to_cpup(p);
502

503 504
	dprintk("lockd: %s status %d state %d\n",
		__func__, resp->status, resp->state);
L
Linus Torvalds 已提交
505 506 507
	return 0;
}

508 509 510
static int nsm_xdr_dec_stat(struct rpc_rqst *rqstp,
			    struct xdr_stream *xdr,
			    struct nsm_res *resp)
L
Linus Torvalds 已提交
511
{
512
	__be32 *p;
513

514
	p = xdr_inline_decode(xdr, 4);
515 516
	if (unlikely(p == NULL))
		return -EIO;
517
	resp->state = be32_to_cpup(p);
518

519
	dprintk("lockd: %s state %d\n", __func__, resp->state);
L
Linus Torvalds 已提交
520 521 522 523
	return 0;
}

#define SM_my_name_sz	(1+XDR_QUADLEN(SM_MAXSTRLEN))
524 525 526
#define SM_my_id_sz	(SM_my_name_sz+3)
#define SM_mon_name_sz	(1+XDR_QUADLEN(SM_MAXSTRLEN))
#define SM_mon_id_sz	(SM_mon_name_sz+SM_my_id_sz)
527 528
#define SM_priv_sz	(XDR_QUADLEN(SM_PRIV_SIZE))
#define SM_mon_sz	(SM_mon_id_sz+SM_priv_sz)
L
Linus Torvalds 已提交
529 530 531 532
#define SM_monres_sz	2
#define SM_unmonres_sz	1

static struct rpc_procinfo	nsm_procedures[] = {
533 534
[NSMPROC_MON] = {
		.p_proc		= NSMPROC_MON,
535
		.p_encode	= (kxdreproc_t)nsm_xdr_enc_mon,
536
		.p_decode	= (kxdrdproc_t)nsm_xdr_dec_stat_res,
537 538
		.p_arglen	= SM_mon_sz,
		.p_replen	= SM_monres_sz,
539
		.p_statidx	= NSMPROC_MON,
540
		.p_name		= "MONITOR",
L
Linus Torvalds 已提交
541
	},
542 543
[NSMPROC_UNMON] = {
		.p_proc		= NSMPROC_UNMON,
544
		.p_encode	= (kxdreproc_t)nsm_xdr_enc_unmon,
545
		.p_decode	= (kxdrdproc_t)nsm_xdr_dec_stat,
546 547
		.p_arglen	= SM_mon_id_sz,
		.p_replen	= SM_unmonres_sz,
548
		.p_statidx	= NSMPROC_UNMON,
549
		.p_name		= "UNMONITOR",
L
Linus Torvalds 已提交
550 551 552
	},
};

553
static const struct rpc_version nsm_version1 = {
554 555
		.number		= 1,
		.nrprocs	= ARRAY_SIZE(nsm_procedures),
L
Linus Torvalds 已提交
556 557 558
		.procs		= nsm_procedures
};

559
static const struct rpc_version *nsm_version[] = {
L
Linus Torvalds 已提交
560 561 562 563 564
	[1] = &nsm_version1,
};

static struct rpc_stat		nsm_stats;

565
static const struct rpc_program nsm_program = {
L
Linus Torvalds 已提交
566
		.name		= "statd",
567
		.number		= NSM_PROGRAM,
568
		.nrvers		= ARRAY_SIZE(nsm_version),
L
Linus Torvalds 已提交
569 570 571
		.version	= nsm_version,
		.stats		= &nsm_stats
};