2 * linux/net/sunrpc/svc.c
4 * High-level RPC service routines
6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
8 * Multiple threads pools and NUMAisation
9 * Copyright (c) 2006 Silicon Graphics, Inc.
10 * by Greg Banks <gnb@melbourne.sgi.com>
13 #include <linux/linkage.h>
14 #include <linux/sched.h>
15 #include <linux/errno.h>
16 #include <linux/net.h>
19 #include <linux/interrupt.h>
20 #include <linux/module.h>
21 #include <linux/kthread.h>
22 #include <linux/slab.h>
23 #include <linux/nsproxy.h>
25 #include <linux/sunrpc/types.h>
26 #include <linux/sunrpc/xdr.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/svcsock.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/sunrpc/bc_xprt.h>
32 #define RPCDBG_FACILITY RPCDBG_SVCDSP
34 static void svc_unregister(const struct svc_serv
*serv
, struct net
*net
);
36 #define svc_serv_is_pooled(serv) ((serv)->sv_function)
39 * Mode for mapping cpus to pools.
42 SVC_POOL_AUTO
= -1, /* choose one of the others */
43 SVC_POOL_GLOBAL
, /* no mapping, just a single global pool
44 * (legacy & UP mode) */
45 SVC_POOL_PERCPU
, /* one pool per cpu */
46 SVC_POOL_PERNODE
/* one pool per numa node */
48 #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
51 * Structure for mapping cpus to pools and vice versa.
52 * Setup once during sunrpc initialisation.
54 static struct svc_pool_map
{
55 int count
; /* How many svc_servs use us */
56 int mode
; /* Note: int not enum to avoid
57 * warnings about "enumeration value
58 * not handled in switch" */
60 unsigned int *pool_to
; /* maps pool id to cpu or node */
61 unsigned int *to_pool
; /* maps cpu or node to pool id */
64 .mode
= SVC_POOL_DEFAULT
66 static DEFINE_MUTEX(svc_pool_map_mutex
);/* protects svc_pool_map.count only */
69 param_set_pool_mode(const char *val
, struct kernel_param
*kp
)
71 int *ip
= (int *)kp
->arg
;
72 struct svc_pool_map
*m
= &svc_pool_map
;
75 mutex_lock(&svc_pool_map_mutex
);
82 if (!strncmp(val
, "auto", 4))
84 else if (!strncmp(val
, "global", 6))
85 *ip
= SVC_POOL_GLOBAL
;
86 else if (!strncmp(val
, "percpu", 6))
87 *ip
= SVC_POOL_PERCPU
;
88 else if (!strncmp(val
, "pernode", 7))
89 *ip
= SVC_POOL_PERNODE
;
94 mutex_unlock(&svc_pool_map_mutex
);
99 param_get_pool_mode(char *buf
, struct kernel_param
*kp
)
101 int *ip
= (int *)kp
->arg
;
106 return strlcpy(buf
, "auto", 20);
107 case SVC_POOL_GLOBAL
:
108 return strlcpy(buf
, "global", 20);
109 case SVC_POOL_PERCPU
:
110 return strlcpy(buf
, "percpu", 20);
111 case SVC_POOL_PERNODE
:
112 return strlcpy(buf
, "pernode", 20);
114 return sprintf(buf
, "%d", *ip
);
118 module_param_call(pool_mode
, param_set_pool_mode
, param_get_pool_mode
,
119 &svc_pool_map
.mode
, 0644);
122 * Detect best pool mapping mode heuristically,
123 * according to the machine's topology.
126 svc_pool_map_choose_mode(void)
130 if (nr_online_nodes
> 1) {
132 * Actually have multiple NUMA nodes,
133 * so split pools on NUMA node boundaries
135 return SVC_POOL_PERNODE
;
138 node
= first_online_node
;
139 if (nr_cpus_node(node
) > 2) {
141 * Non-trivial SMP, or CONFIG_NUMA on
142 * non-NUMA hardware, e.g. with a generic
143 * x86_64 kernel on Xeons. In this case we
144 * want to divide the pools on cpu boundaries.
146 return SVC_POOL_PERCPU
;
149 /* default: one global pool */
150 return SVC_POOL_GLOBAL
;
154 * Allocate the to_pool[] and pool_to[] arrays.
155 * Returns 0 on success or an errno.
158 svc_pool_map_alloc_arrays(struct svc_pool_map
*m
, unsigned int maxpools
)
160 m
->to_pool
= kcalloc(maxpools
, sizeof(unsigned int), GFP_KERNEL
);
163 m
->pool_to
= kcalloc(maxpools
, sizeof(unsigned int), GFP_KERNEL
);
177 * Initialise the pool map for SVC_POOL_PERCPU mode.
178 * Returns number of pools or <0 on error.
181 svc_pool_map_init_percpu(struct svc_pool_map
*m
)
183 unsigned int maxpools
= nr_cpu_ids
;
184 unsigned int pidx
= 0;
188 err
= svc_pool_map_alloc_arrays(m
, maxpools
);
192 for_each_online_cpu(cpu
) {
193 BUG_ON(pidx
> maxpools
);
194 m
->to_pool
[cpu
] = pidx
;
195 m
->pool_to
[pidx
] = cpu
;
198 /* cpus brought online later all get mapped to pool0, sorry */
205 * Initialise the pool map for SVC_POOL_PERNODE mode.
206 * Returns number of pools or <0 on error.
209 svc_pool_map_init_pernode(struct svc_pool_map
*m
)
211 unsigned int maxpools
= nr_node_ids
;
212 unsigned int pidx
= 0;
216 err
= svc_pool_map_alloc_arrays(m
, maxpools
);
220 for_each_node_with_cpus(node
) {
221 /* some architectures (e.g. SN2) have cpuless nodes */
222 BUG_ON(pidx
> maxpools
);
223 m
->to_pool
[node
] = pidx
;
224 m
->pool_to
[pidx
] = node
;
227 /* nodes brought online later all get mapped to pool0, sorry */
234 * Add a reference to the global map of cpus to pools (and
235 * vice versa). Initialise the map if we're the first user.
236 * Returns the number of pools.
239 svc_pool_map_get(void)
241 struct svc_pool_map
*m
= &svc_pool_map
;
244 mutex_lock(&svc_pool_map_mutex
);
247 mutex_unlock(&svc_pool_map_mutex
);
251 if (m
->mode
== SVC_POOL_AUTO
)
252 m
->mode
= svc_pool_map_choose_mode();
255 case SVC_POOL_PERCPU
:
256 npools
= svc_pool_map_init_percpu(m
);
258 case SVC_POOL_PERNODE
:
259 npools
= svc_pool_map_init_pernode(m
);
264 /* default, or memory allocation failure */
266 m
->mode
= SVC_POOL_GLOBAL
;
270 mutex_unlock(&svc_pool_map_mutex
);
276 * Drop a reference to the global map of cpus to pools.
277 * When the last reference is dropped, the map data is
278 * freed; this allows the sysadmin to change the pool
279 * mode using the pool_mode module option without
280 * rebooting or re-loading sunrpc.ko.
283 svc_pool_map_put(void)
285 struct svc_pool_map
*m
= &svc_pool_map
;
287 mutex_lock(&svc_pool_map_mutex
);
297 mutex_unlock(&svc_pool_map_mutex
);
301 static int svc_pool_map_get_node(unsigned int pidx
)
303 const struct svc_pool_map
*m
= &svc_pool_map
;
306 if (m
->mode
== SVC_POOL_PERCPU
)
307 return cpu_to_node(m
->pool_to
[pidx
]);
308 if (m
->mode
== SVC_POOL_PERNODE
)
309 return m
->pool_to
[pidx
];
314 * Set the given thread's cpus_allowed mask so that it
315 * will only run on cpus in the given pool.
318 svc_pool_map_set_cpumask(struct task_struct
*task
, unsigned int pidx
)
320 struct svc_pool_map
*m
= &svc_pool_map
;
321 unsigned int node
= m
->pool_to
[pidx
];
324 * The caller checks for sv_nrpools > 1, which
325 * implies that we've been initialized.
327 BUG_ON(m
->count
== 0);
330 case SVC_POOL_PERCPU
:
332 set_cpus_allowed_ptr(task
, cpumask_of(node
));
335 case SVC_POOL_PERNODE
:
337 set_cpus_allowed_ptr(task
, cpumask_of_node(node
));
344 * Use the mapping mode to choose a pool for a given CPU.
345 * Used when enqueueing an incoming RPC. Always returns
346 * a non-NULL pool pointer.
349 svc_pool_for_cpu(struct svc_serv
*serv
, int cpu
)
351 struct svc_pool_map
*m
= &svc_pool_map
;
352 unsigned int pidx
= 0;
355 * An uninitialised map happens in a pure client when
356 * lockd is brought up, so silently treat it the
357 * same as SVC_POOL_GLOBAL.
359 if (svc_serv_is_pooled(serv
)) {
361 case SVC_POOL_PERCPU
:
362 pidx
= m
->to_pool
[cpu
];
364 case SVC_POOL_PERNODE
:
365 pidx
= m
->to_pool
[cpu_to_node(cpu
)];
369 return &serv
->sv_pools
[pidx
% serv
->sv_nrpools
];
372 static int svc_rpcb_setup(struct svc_serv
*serv
, struct net
*net
)
376 err
= rpcb_create_local(net
);
380 /* Remove any stale portmap registrations */
381 svc_unregister(serv
, net
);
385 void svc_rpcb_cleanup(struct svc_serv
*serv
, struct net
*net
)
387 svc_unregister(serv
, net
);
390 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup
);
392 static int svc_uses_rpcbind(struct svc_serv
*serv
)
394 struct svc_program
*progp
;
397 for (progp
= serv
->sv_program
; progp
; progp
= progp
->pg_next
) {
398 for (i
= 0; i
< progp
->pg_nvers
; i
++) {
399 if (progp
->pg_vers
[i
] == NULL
)
401 if (progp
->pg_vers
[i
]->vs_hidden
== 0)
410 * Create an RPC service
412 static struct svc_serv
*
413 __svc_create(struct svc_program
*prog
, unsigned int bufsize
, int npools
,
414 void (*shutdown
)(struct svc_serv
*serv
, struct net
*net
))
416 struct svc_serv
*serv
;
418 unsigned int xdrsize
;
421 if (!(serv
= kzalloc(sizeof(*serv
), GFP_KERNEL
)))
423 serv
->sv_name
= prog
->pg_name
;
424 serv
->sv_program
= prog
;
425 serv
->sv_nrthreads
= 1;
426 serv
->sv_stats
= prog
->pg_stats
;
427 if (bufsize
> RPCSVC_MAXPAYLOAD
)
428 bufsize
= RPCSVC_MAXPAYLOAD
;
429 serv
->sv_max_payload
= bufsize
? bufsize
: 4096;
430 serv
->sv_max_mesg
= roundup(serv
->sv_max_payload
+ PAGE_SIZE
, PAGE_SIZE
);
431 serv
->sv_shutdown
= shutdown
;
434 prog
->pg_lovers
= prog
->pg_nvers
-1;
435 for (vers
=0; vers
<prog
->pg_nvers
; vers
++)
436 if (prog
->pg_vers
[vers
]) {
437 prog
->pg_hivers
= vers
;
438 if (prog
->pg_lovers
> vers
)
439 prog
->pg_lovers
= vers
;
440 if (prog
->pg_vers
[vers
]->vs_xdrsize
> xdrsize
)
441 xdrsize
= prog
->pg_vers
[vers
]->vs_xdrsize
;
443 prog
= prog
->pg_next
;
445 serv
->sv_xdrsize
= xdrsize
;
446 INIT_LIST_HEAD(&serv
->sv_tempsocks
);
447 INIT_LIST_HEAD(&serv
->sv_permsocks
);
448 init_timer(&serv
->sv_temptimer
);
449 spin_lock_init(&serv
->sv_lock
);
451 serv
->sv_nrpools
= npools
;
453 kcalloc(serv
->sv_nrpools
, sizeof(struct svc_pool
),
455 if (!serv
->sv_pools
) {
460 for (i
= 0; i
< serv
->sv_nrpools
; i
++) {
461 struct svc_pool
*pool
= &serv
->sv_pools
[i
];
463 dprintk("svc: initialising pool %u for %s\n",
467 INIT_LIST_HEAD(&pool
->sp_threads
);
468 INIT_LIST_HEAD(&pool
->sp_sockets
);
469 INIT_LIST_HEAD(&pool
->sp_all_threads
);
470 spin_lock_init(&pool
->sp_lock
);
473 if (svc_uses_rpcbind(serv
)) {
474 if (svc_rpcb_setup(serv
, current
->nsproxy
->net_ns
) < 0) {
475 kfree(serv
->sv_pools
);
479 if (!serv
->sv_shutdown
)
480 serv
->sv_shutdown
= svc_rpcb_cleanup
;
487 svc_create(struct svc_program
*prog
, unsigned int bufsize
,
488 void (*shutdown
)(struct svc_serv
*serv
, struct net
*net
))
490 return __svc_create(prog
, bufsize
, /*npools*/1, shutdown
);
492 EXPORT_SYMBOL_GPL(svc_create
);
495 svc_create_pooled(struct svc_program
*prog
, unsigned int bufsize
,
496 void (*shutdown
)(struct svc_serv
*serv
, struct net
*net
),
497 svc_thread_fn func
, struct module
*mod
)
499 struct svc_serv
*serv
;
500 unsigned int npools
= svc_pool_map_get();
502 serv
= __svc_create(prog
, bufsize
, npools
, shutdown
);
505 serv
->sv_function
= func
;
506 serv
->sv_module
= mod
;
511 EXPORT_SYMBOL_GPL(svc_create_pooled
);
513 void svc_shutdown_net(struct svc_serv
*serv
, struct net
*net
)
516 * The set of xprts (contained in the sv_tempsocks and
517 * sv_permsocks lists) is now constant, since it is modified
518 * only by accepting new sockets (done by service threads in
519 * svc_recv) or aging old ones (done by sv_temptimer), or
520 * configuration changes (excluded by whatever locking the
521 * caller is using--nfsd_mutex in the case of nfsd). So it's
522 * safe to traverse those lists and shut everything down:
524 svc_close_net(serv
, net
);
526 if (serv
->sv_shutdown
)
527 serv
->sv_shutdown(serv
, net
);
529 EXPORT_SYMBOL_GPL(svc_shutdown_net
);
532 * Destroy an RPC service. Should be called with appropriate locking to
533 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
536 svc_destroy(struct svc_serv
*serv
)
538 struct net
*net
= current
->nsproxy
->net_ns
;
540 dprintk("svc: svc_destroy(%s, %d)\n",
541 serv
->sv_program
->pg_name
,
544 if (serv
->sv_nrthreads
) {
545 if (--(serv
->sv_nrthreads
) != 0) {
546 svc_sock_update_bufs(serv
);
550 printk("svc_destroy: no threads for serv=%p!\n", serv
);
552 del_timer_sync(&serv
->sv_temptimer
);
554 svc_shutdown_net(serv
, net
);
557 * The last user is gone and thus all sockets have to be destroyed to
558 * the point. Check this.
560 BUG_ON(!list_empty(&serv
->sv_permsocks
));
561 BUG_ON(!list_empty(&serv
->sv_tempsocks
));
563 cache_clean_deferred(serv
);
565 if (svc_serv_is_pooled(serv
))
568 kfree(serv
->sv_pools
);
571 EXPORT_SYMBOL_GPL(svc_destroy
);
574 * Allocate an RPC server's buffer space.
575 * We allocate pages and place them in rq_argpages.
578 svc_init_buffer(struct svc_rqst
*rqstp
, unsigned int size
, int node
)
580 unsigned int pages
, arghi
;
582 /* bc_xprt uses fore channel allocated buffers */
583 if (svc_is_backchannel(rqstp
))
586 pages
= size
/ PAGE_SIZE
+ 1; /* extra page as we hold both request and reply.
587 * We assume one is at most one page
590 BUG_ON(pages
> RPCSVC_MAXPAGES
);
592 struct page
*p
= alloc_pages_node(node
, GFP_KERNEL
, 0);
595 rqstp
->rq_pages
[arghi
++] = p
;
602 * Release an RPC server buffer
605 svc_release_buffer(struct svc_rqst
*rqstp
)
609 for (i
= 0; i
< ARRAY_SIZE(rqstp
->rq_pages
); i
++)
610 if (rqstp
->rq_pages
[i
])
611 put_page(rqstp
->rq_pages
[i
]);
615 svc_prepare_thread(struct svc_serv
*serv
, struct svc_pool
*pool
, int node
)
617 struct svc_rqst
*rqstp
;
619 rqstp
= kzalloc_node(sizeof(*rqstp
), GFP_KERNEL
, node
);
623 init_waitqueue_head(&rqstp
->rq_wait
);
625 serv
->sv_nrthreads
++;
626 spin_lock_bh(&pool
->sp_lock
);
627 pool
->sp_nrthreads
++;
628 list_add(&rqstp
->rq_all
, &pool
->sp_all_threads
);
629 spin_unlock_bh(&pool
->sp_lock
);
630 rqstp
->rq_server
= serv
;
631 rqstp
->rq_pool
= pool
;
633 rqstp
->rq_argp
= kmalloc_node(serv
->sv_xdrsize
, GFP_KERNEL
, node
);
637 rqstp
->rq_resp
= kmalloc_node(serv
->sv_xdrsize
, GFP_KERNEL
, node
);
641 if (!svc_init_buffer(rqstp
, serv
->sv_max_mesg
, node
))
646 svc_exit_thread(rqstp
);
648 return ERR_PTR(-ENOMEM
);
650 EXPORT_SYMBOL_GPL(svc_prepare_thread
);
653 * Choose a pool in which to create a new thread, for svc_set_num_threads
655 static inline struct svc_pool
*
656 choose_pool(struct svc_serv
*serv
, struct svc_pool
*pool
, unsigned int *state
)
661 return &serv
->sv_pools
[(*state
)++ % serv
->sv_nrpools
];
665 * Choose a thread to kill, for svc_set_num_threads
667 static inline struct task_struct
*
668 choose_victim(struct svc_serv
*serv
, struct svc_pool
*pool
, unsigned int *state
)
671 struct task_struct
*task
= NULL
;
674 spin_lock_bh(&pool
->sp_lock
);
676 /* choose a pool in round-robin fashion */
677 for (i
= 0; i
< serv
->sv_nrpools
; i
++) {
678 pool
= &serv
->sv_pools
[--(*state
) % serv
->sv_nrpools
];
679 spin_lock_bh(&pool
->sp_lock
);
680 if (!list_empty(&pool
->sp_all_threads
))
682 spin_unlock_bh(&pool
->sp_lock
);
688 if (!list_empty(&pool
->sp_all_threads
)) {
689 struct svc_rqst
*rqstp
;
692 * Remove from the pool->sp_all_threads list
693 * so we don't try to kill it again.
695 rqstp
= list_entry(pool
->sp_all_threads
.next
, struct svc_rqst
, rq_all
);
696 list_del_init(&rqstp
->rq_all
);
697 task
= rqstp
->rq_task
;
699 spin_unlock_bh(&pool
->sp_lock
);
705 * Create or destroy enough new threads to make the number
706 * of threads the given number. If `pool' is non-NULL, applies
707 * only to threads in that pool, otherwise round-robins between
708 * all pools. Caller must ensure that mutual exclusion between this and
709 * server startup or shutdown.
711 * Destroying threads relies on the service threads filling in
712 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
713 * has been created using svc_create_pooled().
715 * Based on code that used to be in nfsd_svc() but tweaked
719 svc_set_num_threads(struct svc_serv
*serv
, struct svc_pool
*pool
, int nrservs
)
721 struct svc_rqst
*rqstp
;
722 struct task_struct
*task
;
723 struct svc_pool
*chosen_pool
;
725 unsigned int state
= serv
->sv_nrthreads
-1;
729 /* The -1 assumes caller has done a svc_get() */
730 nrservs
-= (serv
->sv_nrthreads
-1);
732 spin_lock_bh(&pool
->sp_lock
);
733 nrservs
-= pool
->sp_nrthreads
;
734 spin_unlock_bh(&pool
->sp_lock
);
737 /* create new threads */
738 while (nrservs
> 0) {
740 chosen_pool
= choose_pool(serv
, pool
, &state
);
742 node
= svc_pool_map_get_node(chosen_pool
->sp_id
);
743 rqstp
= svc_prepare_thread(serv
, chosen_pool
, node
);
745 error
= PTR_ERR(rqstp
);
749 __module_get(serv
->sv_module
);
750 task
= kthread_create_on_node(serv
->sv_function
, rqstp
,
751 node
, serv
->sv_name
);
753 error
= PTR_ERR(task
);
754 module_put(serv
->sv_module
);
755 svc_exit_thread(rqstp
);
759 rqstp
->rq_task
= task
;
760 if (serv
->sv_nrpools
> 1)
761 svc_pool_map_set_cpumask(task
, chosen_pool
->sp_id
);
763 svc_sock_update_bufs(serv
);
764 wake_up_process(task
);
766 /* destroy old threads */
767 while (nrservs
< 0 &&
768 (task
= choose_victim(serv
, pool
, &state
)) != NULL
) {
769 send_sig(SIGINT
, task
, 1);
775 EXPORT_SYMBOL_GPL(svc_set_num_threads
);
778 * Called from a server thread as it's exiting. Caller must hold the BKL or
779 * the "service mutex", whichever is appropriate for the service.
782 svc_exit_thread(struct svc_rqst
*rqstp
)
784 struct svc_serv
*serv
= rqstp
->rq_server
;
785 struct svc_pool
*pool
= rqstp
->rq_pool
;
787 svc_release_buffer(rqstp
);
788 kfree(rqstp
->rq_resp
);
789 kfree(rqstp
->rq_argp
);
790 kfree(rqstp
->rq_auth_data
);
792 spin_lock_bh(&pool
->sp_lock
);
793 pool
->sp_nrthreads
--;
794 list_del(&rqstp
->rq_all
);
795 spin_unlock_bh(&pool
->sp_lock
);
799 /* Release the server */
803 EXPORT_SYMBOL_GPL(svc_exit_thread
);
806 * Register an "inet" protocol family netid with the local
807 * rpcbind daemon via an rpcbind v4 SET request.
809 * No netconfig infrastructure is available in the kernel, so
810 * we map IP_ protocol numbers to netids by hand.
812 * Returns zero on success; a negative errno value is returned
813 * if any error occurs.
815 static int __svc_rpcb_register4(struct net
*net
, const u32 program
,
817 const unsigned short protocol
,
818 const unsigned short port
)
820 const struct sockaddr_in sin
= {
821 .sin_family
= AF_INET
,
822 .sin_addr
.s_addr
= htonl(INADDR_ANY
),
823 .sin_port
= htons(port
),
830 netid
= RPCBIND_NETID_UDP
;
833 netid
= RPCBIND_NETID_TCP
;
839 error
= rpcb_v4_register(net
, program
, version
,
840 (const struct sockaddr
*)&sin
, netid
);
843 * User space didn't support rpcbind v4, so retry this
844 * registration request with the legacy rpcbind v2 protocol.
846 if (error
== -EPROTONOSUPPORT
)
847 error
= rpcb_register(net
, program
, version
, protocol
, port
);
852 #if IS_ENABLED(CONFIG_IPV6)
854 * Register an "inet6" protocol family netid with the local
855 * rpcbind daemon via an rpcbind v4 SET request.
857 * No netconfig infrastructure is available in the kernel, so
858 * we map IP_ protocol numbers to netids by hand.
860 * Returns zero on success; a negative errno value is returned
861 * if any error occurs.
863 static int __svc_rpcb_register6(struct net
*net
, const u32 program
,
865 const unsigned short protocol
,
866 const unsigned short port
)
868 const struct sockaddr_in6 sin6
= {
869 .sin6_family
= AF_INET6
,
870 .sin6_addr
= IN6ADDR_ANY_INIT
,
871 .sin6_port
= htons(port
),
878 netid
= RPCBIND_NETID_UDP6
;
881 netid
= RPCBIND_NETID_TCP6
;
887 error
= rpcb_v4_register(net
, program
, version
,
888 (const struct sockaddr
*)&sin6
, netid
);
891 * User space didn't support rpcbind version 4, so we won't
892 * use a PF_INET6 listener.
894 if (error
== -EPROTONOSUPPORT
)
895 error
= -EAFNOSUPPORT
;
899 #endif /* IS_ENABLED(CONFIG_IPV6) */
902 * Register a kernel RPC service via rpcbind version 4.
904 * Returns zero on success; a negative errno value is returned
905 * if any error occurs.
907 static int __svc_register(struct net
*net
, const char *progname
,
908 const u32 program
, const u32 version
,
910 const unsigned short protocol
,
911 const unsigned short port
)
913 int error
= -EAFNOSUPPORT
;
917 error
= __svc_rpcb_register4(net
, program
, version
,
920 #if IS_ENABLED(CONFIG_IPV6)
922 error
= __svc_rpcb_register6(net
, program
, version
,
928 printk(KERN_WARNING
"svc: failed to register %sv%u RPC "
929 "service (errno %d).\n", progname
, version
, -error
);
934 * svc_register - register an RPC service with the local portmapper
935 * @serv: svc_serv struct for the service to register
936 * @net: net namespace for the service to register
937 * @family: protocol family of service's listener socket
938 * @proto: transport protocol number to advertise
939 * @port: port to advertise
941 * Service is registered for any address in the passed-in protocol family
943 int svc_register(const struct svc_serv
*serv
, struct net
*net
,
944 const int family
, const unsigned short proto
,
945 const unsigned short port
)
947 struct svc_program
*progp
;
951 BUG_ON(proto
== 0 && port
== 0);
953 for (progp
= serv
->sv_program
; progp
; progp
= progp
->pg_next
) {
954 for (i
= 0; i
< progp
->pg_nvers
; i
++) {
955 if (progp
->pg_vers
[i
] == NULL
)
958 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
961 proto
== IPPROTO_UDP
? "udp" : "tcp",
964 progp
->pg_vers
[i
]->vs_hidden
?
965 " (but not telling portmap)" : "");
967 if (progp
->pg_vers
[i
]->vs_hidden
)
970 error
= __svc_register(net
, progp
->pg_name
, progp
->pg_prog
,
971 i
, family
, proto
, port
);
981 * If user space is running rpcbind, it should take the v4 UNSET
982 * and clear everything for this [program, version]. If user space
983 * is running portmap, it will reject the v4 UNSET, but won't have
984 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
985 * in this case to clear all existing entries for [program, version].
987 static void __svc_unregister(struct net
*net
, const u32 program
, const u32 version
,
988 const char *progname
)
992 error
= rpcb_v4_register(net
, program
, version
, NULL
, "");
995 * User space didn't support rpcbind v4, so retry this
996 * request with the legacy rpcbind v2 protocol.
998 if (error
== -EPROTONOSUPPORT
)
999 error
= rpcb_register(net
, program
, version
, 0, 0);
1001 dprintk("svc: %s(%sv%u), error %d\n",
1002 __func__
, progname
, version
, error
);
1006 * All netids, bind addresses and ports registered for [program, version]
1007 * are removed from the local rpcbind database (if the service is not
1008 * hidden) to make way for a new instance of the service.
1010 * The result of unregistration is reported via dprintk for those who want
1011 * verification of the result, but is otherwise not important.
1013 static void svc_unregister(const struct svc_serv
*serv
, struct net
*net
)
1015 struct svc_program
*progp
;
1016 unsigned long flags
;
1019 clear_thread_flag(TIF_SIGPENDING
);
1021 for (progp
= serv
->sv_program
; progp
; progp
= progp
->pg_next
) {
1022 for (i
= 0; i
< progp
->pg_nvers
; i
++) {
1023 if (progp
->pg_vers
[i
] == NULL
)
1025 if (progp
->pg_vers
[i
]->vs_hidden
)
1028 dprintk("svc: attempting to unregister %sv%u\n",
1030 __svc_unregister(net
, progp
->pg_prog
, i
, progp
->pg_name
);
1034 spin_lock_irqsave(¤t
->sighand
->siglock
, flags
);
1035 recalc_sigpending();
1036 spin_unlock_irqrestore(¤t
->sighand
->siglock
, flags
);
1040 * Printk the given error with the address of the client that caused it.
1042 static __printf(2, 3)
1043 int svc_printk(struct svc_rqst
*rqstp
, const char *fmt
, ...)
1047 char buf
[RPC_MAX_ADDRBUFLEN
];
1049 if (!net_ratelimit())
1052 printk(KERN_WARNING
"svc: %s: ",
1053 svc_print_addr(rqstp
, buf
, sizeof(buf
)));
1055 va_start(args
, fmt
);
1056 r
= vprintk(fmt
, args
);
1063 * Common routine for processing the RPC request.
1066 svc_process_common(struct svc_rqst
*rqstp
, struct kvec
*argv
, struct kvec
*resv
)
1068 struct svc_program
*progp
;
1069 struct svc_version
*versp
= NULL
; /* compiler food */
1070 struct svc_procedure
*procp
= NULL
;
1071 struct svc_serv
*serv
= rqstp
->rq_server
;
1074 u32 prog
, vers
, proc
;
1075 __be32 auth_stat
, rpc_stat
;
1077 __be32
*reply_statp
;
1079 rpc_stat
= rpc_success
;
1081 if (argv
->iov_len
< 6*4)
1084 /* Will be turned off only in gss privacy case: */
1085 rqstp
->rq_splice_ok
= 1;
1086 /* Will be turned off only when NFSv4 Sessions are used */
1087 rqstp
->rq_usedeferral
= 1;
1088 rqstp
->rq_dropme
= false;
1090 /* Setup reply header */
1091 rqstp
->rq_xprt
->xpt_ops
->xpo_prep_reply_hdr(rqstp
);
1093 svc_putu32(resv
, rqstp
->rq_xid
);
1095 vers
= svc_getnl(argv
);
1097 /* First words of reply: */
1098 svc_putnl(resv
, 1); /* REPLY */
1100 if (vers
!= 2) /* RPC version number */
1103 /* Save position in case we later decide to reject: */
1104 reply_statp
= resv
->iov_base
+ resv
->iov_len
;
1106 svc_putnl(resv
, 0); /* ACCEPT */
1108 rqstp
->rq_prog
= prog
= svc_getnl(argv
); /* program number */
1109 rqstp
->rq_vers
= vers
= svc_getnl(argv
); /* version number */
1110 rqstp
->rq_proc
= proc
= svc_getnl(argv
); /* procedure number */
1112 progp
= serv
->sv_program
;
1114 for (progp
= serv
->sv_program
; progp
; progp
= progp
->pg_next
)
1115 if (prog
== progp
->pg_prog
)
1119 * Decode auth data, and add verifier to reply buffer.
1120 * We do this before anything else in order to get a decent
1123 auth_res
= svc_authenticate(rqstp
, &auth_stat
);
1124 /* Also give the program a chance to reject this call: */
1125 if (auth_res
== SVC_OK
&& progp
) {
1126 auth_stat
= rpc_autherr_badcred
;
1127 auth_res
= progp
->pg_authenticate(rqstp
);
1135 rpc_stat
= rpc_system_err
;
1140 if (test_bit(XPT_TEMP
, &rqstp
->rq_xprt
->xpt_flags
))
1141 svc_close_xprt(rqstp
->rq_xprt
);
1151 if (vers
>= progp
->pg_nvers
||
1152 !(versp
= progp
->pg_vers
[vers
]))
1155 procp
= versp
->vs_proc
+ proc
;
1156 if (proc
>= versp
->vs_nproc
|| !procp
->pc_func
)
1158 rqstp
->rq_procinfo
= procp
;
1160 /* Syntactic check complete */
1161 serv
->sv_stats
->rpccnt
++;
1163 /* Build the reply header. */
1164 statp
= resv
->iov_base
+resv
->iov_len
;
1165 svc_putnl(resv
, RPC_SUCCESS
);
1167 /* Bump per-procedure stats counter */
1170 /* Initialize storage for argp and resp */
1171 memset(rqstp
->rq_argp
, 0, procp
->pc_argsize
);
1172 memset(rqstp
->rq_resp
, 0, procp
->pc_ressize
);
1174 /* un-reserve some of the out-queue now that we have a
1175 * better idea of reply size
1177 if (procp
->pc_xdrressize
)
1178 svc_reserve_auth(rqstp
, procp
->pc_xdrressize
<<2);
1180 /* Call the function that processes the request. */
1181 if (!versp
->vs_dispatch
) {
1182 /* Decode arguments */
1183 xdr
= procp
->pc_decode
;
1184 if (xdr
&& !xdr(rqstp
, argv
->iov_base
, rqstp
->rq_argp
))
1187 *statp
= procp
->pc_func(rqstp
, rqstp
->rq_argp
, rqstp
->rq_resp
);
1190 if (rqstp
->rq_dropme
) {
1191 if (procp
->pc_release
)
1192 procp
->pc_release(rqstp
, NULL
, rqstp
->rq_resp
);
1195 if (*statp
== rpc_success
&&
1196 (xdr
= procp
->pc_encode
) &&
1197 !xdr(rqstp
, resv
->iov_base
+resv
->iov_len
, rqstp
->rq_resp
)) {
1198 dprintk("svc: failed to encode reply\n");
1199 /* serv->sv_stats->rpcsystemerr++; */
1200 *statp
= rpc_system_err
;
1203 dprintk("svc: calling dispatcher\n");
1204 if (!versp
->vs_dispatch(rqstp
, statp
)) {
1205 /* Release reply info */
1206 if (procp
->pc_release
)
1207 procp
->pc_release(rqstp
, NULL
, rqstp
->rq_resp
);
1212 /* Check RPC status result */
1213 if (*statp
!= rpc_success
)
1214 resv
->iov_len
= ((void*)statp
) - resv
->iov_base
+ 4;
1216 /* Release reply info */
1217 if (procp
->pc_release
)
1218 procp
->pc_release(rqstp
, NULL
, rqstp
->rq_resp
);
1220 if (procp
->pc_encode
== NULL
)
1224 if (svc_authorise(rqstp
))
1226 return 1; /* Caller can now send it */
1229 svc_authorise(rqstp
); /* doesn't hurt to call this twice */
1230 dprintk("svc: svc_process dropit\n");
1234 svc_printk(rqstp
, "short len %Zd, dropping request\n",
1237 goto dropit
; /* drop request */
1240 serv
->sv_stats
->rpcbadfmt
++;
1241 svc_putnl(resv
, 1); /* REJECT */
1242 svc_putnl(resv
, 0); /* RPC_MISMATCH */
1243 svc_putnl(resv
, 2); /* Only RPCv2 supported */
1248 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat
));
1249 serv
->sv_stats
->rpcbadauth
++;
1250 /* Restore write pointer to location of accept status: */
1251 xdr_ressize_check(rqstp
, reply_statp
);
1252 svc_putnl(resv
, 1); /* REJECT */
1253 svc_putnl(resv
, 1); /* AUTH_ERROR */
1254 svc_putnl(resv
, ntohl(auth_stat
)); /* status */
1258 dprintk("svc: unknown program %d\n", prog
);
1259 serv
->sv_stats
->rpcbadfmt
++;
1260 svc_putnl(resv
, RPC_PROG_UNAVAIL
);
1264 svc_printk(rqstp
, "unknown version (%d for prog %d, %s)\n",
1265 vers
, prog
, progp
->pg_name
);
1267 serv
->sv_stats
->rpcbadfmt
++;
1268 svc_putnl(resv
, RPC_PROG_MISMATCH
);
1269 svc_putnl(resv
, progp
->pg_lovers
);
1270 svc_putnl(resv
, progp
->pg_hivers
);
1274 svc_printk(rqstp
, "unknown procedure (%d)\n", proc
);
1276 serv
->sv_stats
->rpcbadfmt
++;
1277 svc_putnl(resv
, RPC_PROC_UNAVAIL
);
1281 svc_printk(rqstp
, "failed to decode args\n");
1283 rpc_stat
= rpc_garbage_args
;
1285 serv
->sv_stats
->rpcbadfmt
++;
1286 svc_putnl(resv
, ntohl(rpc_stat
));
1289 EXPORT_SYMBOL_GPL(svc_process
);
1292 * Process the RPC request.
1295 svc_process(struct svc_rqst
*rqstp
)
1297 struct kvec
*argv
= &rqstp
->rq_arg
.head
[0];
1298 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1299 struct svc_serv
*serv
= rqstp
->rq_server
;
1303 * Setup response xdr_buf.
1304 * Initially it has just one page
1306 rqstp
->rq_resused
= 1;
1307 resv
->iov_base
= page_address(rqstp
->rq_respages
[0]);
1309 rqstp
->rq_res
.pages
= rqstp
->rq_respages
+ 1;
1310 rqstp
->rq_res
.len
= 0;
1311 rqstp
->rq_res
.page_base
= 0;
1312 rqstp
->rq_res
.page_len
= 0;
1313 rqstp
->rq_res
.buflen
= PAGE_SIZE
;
1314 rqstp
->rq_res
.tail
[0].iov_base
= NULL
;
1315 rqstp
->rq_res
.tail
[0].iov_len
= 0;
1317 rqstp
->rq_xid
= svc_getu32(argv
);
1319 dir
= svc_getnl(argv
);
1321 /* direction != CALL */
1322 svc_printk(rqstp
, "bad direction %d, dropping request\n", dir
);
1323 serv
->sv_stats
->rpcbadfmt
++;
1328 /* Returns 1 for send, 0 for drop */
1329 if (svc_process_common(rqstp
, argv
, resv
))
1330 return svc_send(rqstp
);
1337 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1339 * Process a backchannel RPC request that arrived over an existing
1340 * outbound connection
1343 bc_svc_process(struct svc_serv
*serv
, struct rpc_rqst
*req
,
1344 struct svc_rqst
*rqstp
)
1346 struct kvec
*argv
= &rqstp
->rq_arg
.head
[0];
1347 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1349 /* Build the svc_rqst used by the common processing routine */
1350 rqstp
->rq_xprt
= serv
->sv_bc_xprt
;
1351 rqstp
->rq_xid
= req
->rq_xid
;
1352 rqstp
->rq_prot
= req
->rq_xprt
->prot
;
1353 rqstp
->rq_server
= serv
;
1355 rqstp
->rq_addrlen
= sizeof(req
->rq_xprt
->addr
);
1356 memcpy(&rqstp
->rq_addr
, &req
->rq_xprt
->addr
, rqstp
->rq_addrlen
);
1357 memcpy(&rqstp
->rq_arg
, &req
->rq_rcv_buf
, sizeof(rqstp
->rq_arg
));
1358 memcpy(&rqstp
->rq_res
, &req
->rq_snd_buf
, sizeof(rqstp
->rq_res
));
1360 /* reset result send buffer "put" position */
1363 if (rqstp
->rq_prot
!= IPPROTO_TCP
) {
1364 printk(KERN_ERR
"No support for Non-TCP transports!\n");
1369 * Skip the next two words because they've already been
1370 * processed in the trasport
1372 svc_getu32(argv
); /* XID */
1373 svc_getnl(argv
); /* CALLDIR */
1375 /* Returns 1 for send, 0 for drop */
1376 if (svc_process_common(rqstp
, argv
, resv
)) {
1377 memcpy(&req
->rq_snd_buf
, &rqstp
->rq_res
,
1378 sizeof(req
->rq_snd_buf
));
1379 return bc_send(req
);
1381 /* Nothing to do to drop request */
1385 EXPORT_SYMBOL_GPL(bc_svc_process
);
1386 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1389 * Return (transport-specific) limit on the rpc payload.
1391 u32
svc_max_payload(const struct svc_rqst
*rqstp
)
1393 u32 max
= rqstp
->rq_xprt
->xpt_class
->xcl_max_payload
;
1395 if (rqstp
->rq_server
->sv_max_payload
< max
)
1396 max
= rqstp
->rq_server
->sv_max_payload
;
1399 EXPORT_SYMBOL_GPL(svc_max_payload
);